data/sae/index.json

{
 "sae": {
  "sae_id": "full_650m_L035_x8_k64",
  "model_id": "cherry_idx27_att1@200000",
  "model_name": "Botanic1-M",
  "layer": 35,
  "d_sae": 8192,
  "k": 64,
  "expansion": 8,
  "hook_site": "resid_post",
  "sae_threshold": 0.8141039609909058,
  "sae_sha256": "c40384e655b3135c9758eeaea654b14d0d9617ccf11e5e33b6e16e203d5cac6c",
  "threshold_mode": "pinned",
  "note": "Botanic1-M, layer 35, x8 expansion, k = 64"
 },
 "curated_features": {
  "4131": {
   "label": "splice donor",
   "kind": "site",
   "evidence": "MCC 0.871 at a validation-selected threshold; precision 0.876, recall 0.866 at prevalence 6.4e-4"
  },
  "2634": {
   "label": "splice donor, usage-graded",
   "kind": "site",
   "evidence": "Spearman rho 0.759 against percent splice-site usage over 3,254 annotated donors"
  },
  "586": {
   "label": "splice acceptor",
   "kind": "site",
   "evidence": "MCC 0.873; precision 0.845, recall 0.901 at prevalence 6.3e-4"
  },
  "7708": {
   "label": "intron region",
   "kind": "range",
   "evidence": "MCC 0.694"
  },
  "3930": {
   "label": "CDS codon position 1",
   "kind": "range",
   "evidence": "MCC 0.668"
  },
  "2618": {
   "label": "CDS codon position 2",
   "kind": "range",
   "evidence": "MCC 0.542"
  },
  "1200": {
   "label": "CDS codon position 3",
   "kind": "range",
   "evidence": "MCC 0.633"
  },
  "4567": {
   "label": "W-box, + strand",
   "kind": "motif",
   "evidence": "TTGAC core in 30 of the feature's 32 top firings; no motif database used"
  },
  "145": {
   "label": "W-box, - strand",
   "kind": "motif",
   "evidence": "GTCAA core in 28 of the feature's 32 top firings"
  },
  "5804": {
   "label": "G-box",
   "kind": "motif",
   "evidence": "CACGTG core in 17 of the feature's 32 top firings"
  },
  "1474": {
   "label": "tandem repeat",
   "kind": "pattern",
   "evidence": "one dip per repeat unit in top firings"
  },
  "5349": {
   "label": "microsatellite (AT)n",
   "kind": "pattern",
   "evidence": "one base in two across an (AT)n tract"
  }
 },
 "windows": [
  {
   "window_uid": "AT1G65170",
   "featured": true,
   "species": "athaliana",
   "species_binomial": "Arabidopsis thaliana",
   "family": "Brassicaceae",
   "order": "Brassicales",
   "assembly": "TAIR10",
   "chrom": "1",
   "start": 24206269,
   "end": 24214460,
   "strand": 1,
   "n_bases": 8191,
   "split": null,
   "anchor": {
    "kind": "donor",
    "offset": 3857
   },
   "note": "The locus the report's annotation-correction result is drawn from. Both donor features read exactly 0.00 at the donor TAIR10 and Araport11 annotate, and fire 54 bp downstream at the site the 1999 BAC submission and the 2026 TAIR12 release both place.",
   "genes_sense": [
    "AT1G65160",
    "AT1G65165",
    "AT1G65170"
   ],
   "n_features": 1292,
   "default_features": [
    4131,
    2634,
    38,
    5594,
    3797,
    271,
    266,
    1200,
    7929
   ],
   "lane_rationale": [
    {
     "id": 4131,
     "marks": "the claim this window is here for",
     "nice": "the contested site",
     "mcc_in_window": null,
     "cut_percentile": null
    },
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.663,
     "cut_percentile": 40
    },
    {
     "id": 38,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.621,
     "cut_percentile": 60
    },
    {
     "id": 5594,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.725,
     "cut_percentile": 40
    },
    {
     "id": 3797,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.479,
     "cut_percentile": 40
    },
    {
     "id": 271,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.777,
     "cut_percentile": 40
    },
    {
     "id": 266,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.617,
     "cut_percentile": 40
    },
    {
     "id": 1200,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.636,
     "cut_percentile": 40
    },
    {
     "id": 7929,
     "marks": "antisense:utr5",
     "nice": "antisense 5\u2032 UTR",
     "mcc_in_window": 0.797,
     "cut_percentile": 60
    }
   ],
   "coverage": {
    "features_stored": 1294,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 4,
   "n_sites": 43,
   "bytes": 2413489
  },
  {
   "window_uid": "B00001890",
   "featured": false,
   "species": "athaliana",
   "species_binomial": "Arabidopsis thaliana",
   "family": "Brassicaceae",
   "order": "Brassicales",
   "assembly": "TAIR10",
   "chrom": "5",
   "start": 16782652,
   "end": 16790843,
   "strand": 1,
   "n_bases": 8191,
   "split": "final_test",
   "anchor": {
    "kind": "tis",
    "offset": 3172
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 43 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.95, and 100% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "AT5G41940",
    "AT5G41950",
    "AT5G41960"
   ],
   "n_features": 1117,
   "default_features": [
    2634,
    586,
    7708,
    7224,
    3160,
    2584,
    2425,
    5782
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.809,
     "cut_percentile": 40
    },
    {
     "id": 586,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.677,
     "cut_percentile": 60
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.674,
     "cut_percentile": 40
    },
    {
     "id": 7224,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.406,
     "cut_percentile": 40
    },
    {
     "id": 3160,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.594,
     "cut_percentile": 40
    },
    {
     "id": 2584,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.514,
     "cut_percentile": 40
    },
    {
     "id": 2425,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.784,
     "cut_percentile": 75
    },
    {
     "id": 5782,
     "marks": "sense:utr3",
     "nice": "3\u2032 UTR",
     "mcc_in_window": 0.699,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1119,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 3,
   "n_sites": 51,
   "bytes": 2400942,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 43 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.95, and 100% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.981
   }
  },
  {
   "window_uid": "B00002447",
   "featured": false,
   "species": "athaliana",
   "species_binomial": "Arabidopsis thaliana",
   "family": "Brassicaceae",
   "order": "Brassicales",
   "assembly": "TAIR10",
   "chrom": "3",
   "start": 3619444,
   "end": 3627635,
   "strand": -1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "tis",
    "offset": 2769
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 24 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.95, and 100% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "AT3G11490",
    "AT3G11500",
    "AT3G11505"
   ],
   "n_features": 1299,
   "default_features": [
    2634,
    2993,
    7708,
    7944,
    2798,
    7224,
    2425,
    3051
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.761,
     "cut_percentile": 60
    },
    {
     "id": 2993,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.629,
     "cut_percentile": 60
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.74,
     "cut_percentile": 40
    },
    {
     "id": 7944,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.49,
     "cut_percentile": 40
    },
    {
     "id": 2798,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.755,
     "cut_percentile": 40
    },
    {
     "id": 7224,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.806,
     "cut_percentile": 60
    },
    {
     "id": 2425,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.816,
     "cut_percentile": 88
    },
    {
     "id": 3051,
     "marks": "sense:ncrna",
     "nice": "non-coding exon",
     "mcc_in_window": 0.947,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1301,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 6,
   "n_sites": 38,
   "bytes": 2518110,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 24 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.95, and 100% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.982
   }
  },
  {
   "window_uid": "B00011941",
   "featured": false,
   "species": "osativa",
   "species_binomial": "Oryza sativa",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "IRGSP-1.0",
   "chrom": "2",
   "start": 23865777,
   "end": 23873968,
   "strand": -1,
   "n_bases": 8191,
   "split": "interp_val",
   "anchor": {
    "kind": "donor",
    "offset": 3992
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 27 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.98, and 99% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "Os02g0608400"
   ],
   "n_features": 1291,
   "default_features": [
    2634,
    5569,
    5515,
    5811,
    1564,
    4764,
    2467,
    7056
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.809,
     "cut_percentile": 60
    },
    {
     "id": 5569,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.708,
     "cut_percentile": 75
    },
    {
     "id": 5515,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.64,
     "cut_percentile": 40
    },
    {
     "id": 5811,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.551,
     "cut_percentile": 40
    },
    {
     "id": 1564,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.717,
     "cut_percentile": 40
    },
    {
     "id": 4764,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.615,
     "cut_percentile": 40
    },
    {
     "id": 2467,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.738,
     "cut_percentile": 40
    },
    {
     "id": 7056,
     "marks": "sense:utr5",
     "nice": "5\u2032 UTR",
     "mcc_in_window": 0.976,
     "cut_percentile": 60
    }
   ],
   "coverage": {
    "features_stored": 1292,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 30,
   "bytes": 2290925,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 27 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.98, and 99% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.989
   }
  },
  {
   "window_uid": "B00012077",
   "featured": false,
   "species": "osativa",
   "species_binomial": "Oryza sativa",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "IRGSP-1.0",
   "chrom": "2",
   "start": 32125877,
   "end": 32134068,
   "strand": -1,
   "n_bases": 8191,
   "split": "interp_val",
   "anchor": {
    "kind": "donor",
    "offset": 4786
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 39 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 99% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "OsRad54"
   ],
   "n_features": 1338,
   "default_features": [
    2634,
    586,
    7708,
    7447,
    80,
    49,
    2425,
    5957
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.799,
     "cut_percentile": 60
    },
    {
     "id": 586,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.687,
     "cut_percentile": 40
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.568,
     "cut_percentile": 40
    },
    {
     "id": 7447,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.52,
     "cut_percentile": 40
    },
    {
     "id": 80,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.754,
     "cut_percentile": 40
    },
    {
     "id": 49,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.732,
     "cut_percentile": 40
    },
    {
     "id": 2425,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.909,
     "cut_percentile": 75
    },
    {
     "id": 5957,
     "marks": "sense:utr3",
     "nice": "3\u2032 UTR",
     "mcc_in_window": 0.861,
     "cut_percentile": 60
    }
   ],
   "coverage": {
    "features_stored": 1338,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 40,
   "bytes": 2366784,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 39 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 99% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.986
   }
  },
  {
   "window_uid": "B00014101",
   "featured": false,
   "species": "osativa",
   "species_binomial": "Oryza sativa",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "IRGSP-1.0",
   "chrom": "11",
   "start": 18381867,
   "end": 18390058,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_val",
   "anchor": {
    "kind": "acceptor",
    "offset": 4280
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 47 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 98% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "Os11g0513900"
   ],
   "n_features": 1291,
   "default_features": [
    2634,
    38,
    7708,
    1668,
    80,
    3666,
    2425,
    2325
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.837,
     "cut_percentile": 40
    },
    {
     "id": 38,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.718,
     "cut_percentile": 75
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.624,
     "cut_percentile": 40
    },
    {
     "id": 1668,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.596,
     "cut_percentile": 40
    },
    {
     "id": 80,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.746,
     "cut_percentile": 40
    },
    {
     "id": 3666,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.737,
     "cut_percentile": 40
    },
    {
     "id": 2425,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.822,
     "cut_percentile": 75
    },
    {
     "id": 2325,
     "marks": "sense:utr5",
     "nice": "5\u2032 UTR",
     "mcc_in_window": 0.674,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1292,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 49,
   "bytes": 2379539,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 47 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 98% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.982
   }
  },
  {
   "window_uid": "B00026784",
   "featured": false,
   "species": "slycopersicum",
   "species_binomial": "Solanum lycopersicum",
   "family": null,
   "order": null,
   "assembly": "SL3.0",
   "chrom": "7",
   "start": 64571128,
   "end": 64579319,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_val",
   "anchor": {
    "kind": "tis",
    "offset": 1553
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 16 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.96, and 100% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "Solyc07g056630.3"
   ],
   "n_features": 1559,
   "default_features": [
    2634,
    586,
    7708,
    7447,
    3160,
    1468,
    2425,
    520
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.757,
     "cut_percentile": 40
    },
    {
     "id": 586,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.622,
     "cut_percentile": 75
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.563,
     "cut_percentile": 40
    },
    {
     "id": 7447,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.549,
     "cut_percentile": 40
    },
    {
     "id": 3160,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.739,
     "cut_percentile": 40
    },
    {
     "id": 1468,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.692,
     "cut_percentile": 40
    },
    {
     "id": 2425,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.866,
     "cut_percentile": 75
    },
    {
     "id": 520,
     "marks": "sense:utr5",
     "nice": "5\u2032 UTR",
     "mcc_in_window": 0.61,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1560,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 17,
   "bytes": 2105316,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 16 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.96, and 100% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.988
   }
  },
  {
   "window_uid": "B00027277",
   "featured": false,
   "species": "slycopersicum",
   "species_binomial": "Solanum lycopersicum",
   "family": null,
   "order": null,
   "assembly": "SL3.0",
   "chrom": "6",
   "start": 42400943,
   "end": 42409134,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "donor",
    "offset": 6718
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 43 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 100% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "Solyc06g068230.3"
   ],
   "n_features": 1354,
   "default_features": [
    2634,
    586,
    7708,
    2467,
    80,
    49,
    2425,
    7347
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.768,
     "cut_percentile": 40
    },
    {
     "id": 586,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.748,
     "cut_percentile": 60
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.638,
     "cut_percentile": 40
    },
    {
     "id": 2467,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.446,
     "cut_percentile": 40
    },
    {
     "id": 80,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.748,
     "cut_percentile": 40
    },
    {
     "id": 49,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.634,
     "cut_percentile": 40
    },
    {
     "id": 2425,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.808,
     "cut_percentile": 75
    },
    {
     "id": 7347,
     "marks": "sense:utr5",
     "nice": "5\u2032 UTR",
     "mcc_in_window": 0.801,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1355,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 44,
   "bytes": 2410186,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 43 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 100% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.99
   }
  },
  {
   "window_uid": "B00030192",
   "featured": false,
   "species": "slycopersicum",
   "species_binomial": "Solanum lycopersicum",
   "family": null,
   "order": null,
   "assembly": "SL3.0",
   "chrom": "9",
   "start": 5569386,
   "end": 5577577,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "acceptor",
    "offset": 5397
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 28 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.98, and 100% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "Solyc09g013160.3"
   ],
   "n_features": 1463,
   "default_features": [
    2634,
    5569,
    7708,
    2360,
    2230,
    2802,
    2425,
    6059
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.809,
     "cut_percentile": 40
    },
    {
     "id": 5569,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.657,
     "cut_percentile": 88
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.674,
     "cut_percentile": 40
    },
    {
     "id": 2360,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.534,
     "cut_percentile": 40
    },
    {
     "id": 2230,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.749,
     "cut_percentile": 40
    },
    {
     "id": 2802,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.885,
     "cut_percentile": 40
    },
    {
     "id": 2425,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.906,
     "cut_percentile": 75
    },
    {
     "id": 6059,
     "marks": "sense:utr5",
     "nice": "5\u2032 UTR",
     "mcc_in_window": 0.699,
     "cut_percentile": 60
    }
   ],
   "coverage": {
    "features_stored": 1464,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 30,
   "bytes": 2290930,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 28 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.98, and 100% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.992
   }
  },
  {
   "window_uid": "B00019012",
   "featured": false,
   "species": "zmays",
   "species_binomial": "Zea mays",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "Zm-B73-REFERENCE-NAM-5.0",
   "chrom": "10",
   "start": 125920467,
   "end": 125928658,
   "strand": -1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "tis",
    "offset": 1953
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 28 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 99% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "Zm00001eb424400",
    "AMADH2"
   ],
   "n_features": 1605,
   "default_features": [
    2634,
    2993,
    7708,
    7447,
    3784,
    7275,
    2425,
    5782
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.794,
     "cut_percentile": 60
    },
    {
     "id": 2993,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.632,
     "cut_percentile": 60
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.69,
     "cut_percentile": 40
    },
    {
     "id": 7447,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.433,
     "cut_percentile": 40
    },
    {
     "id": 3784,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.683,
     "cut_percentile": 40
    },
    {
     "id": 7275,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.544,
     "cut_percentile": 40
    },
    {
     "id": 2425,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.702,
     "cut_percentile": 88
    },
    {
     "id": 5782,
     "marks": "sense:utr3",
     "nice": "3\u2032 UTR",
     "mcc_in_window": 0.9,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1607,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 2,
   "n_sites": 34,
   "bytes": 2490776,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 28 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 99% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.989
   }
  },
  {
   "window_uid": "B00021698",
   "featured": false,
   "species": "zmays",
   "species_binomial": "Zea mays",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "Zm-B73-REFERENCE-NAM-5.0",
   "chrom": "2",
   "start": 241877090,
   "end": 241885281,
   "strand": -1,
   "n_bases": 8191,
   "split": "interp_val",
   "anchor": {
    "kind": "acceptor",
    "offset": 7015
   },
   "note": "features track the annotation \u2014 the splice features fire at 100% of the 20 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 99% of the codon-position feature's firing falls inside coding sequence",
   "genes_sense": [
    "MAPKKK5_0"
   ],
   "n_features": 1564,
   "default_features": [
    2634,
    2993,
    5515,
    3267,
    2655,
    3333,
    2301,
    5782
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.807,
     "cut_percentile": 60
    },
    {
     "id": 2993,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.632,
     "cut_percentile": 40
    },
    {
     "id": 5515,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.697,
     "cut_percentile": 40
    },
    {
     "id": 3267,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.638,
     "cut_percentile": 40
    },
    {
     "id": 2655,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.672,
     "cut_percentile": 40
    },
    {
     "id": 3333,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.654,
     "cut_percentile": 40
    },
    {
     "id": 2301,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.702,
     "cut_percentile": 40
    },
    {
     "id": 5782,
     "marks": "sense:utr3",
     "nice": "3\u2032 UTR",
     "mcc_in_window": 0.914,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1566,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 24,
   "bytes": 2416793,
   "selection": {
    "kind": "agreement",
    "headline": "features track the annotation",
    "why": "the splice features fire at 100% of the 20 annotated donors and acceptors, the intron feature overlaps the annotated introns at Jaccard 0.97, and 99% of the codon-position feature's firing falls inside coding sequence",
    "score": 0.987
   }
  },
  {
   "window_uid": "B00003566",
   "featured": false,
   "species": "athaliana",
   "species_binomial": "Arabidopsis thaliana",
   "family": "Brassicaceae",
   "order": "Brassicales",
   "assembly": "TAIR10",
   "chrom": "3",
   "start": 787414,
   "end": 795605,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "donor",
    "offset": 4145
   },
   "note": "an alternative splice site the canonical track drops \u2014 f586 reads 15.9 at chr3:793,413 on the sense strand, 38 bp from the canonical acceptor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads AG",
   "genes_sense": [
    "UNE6",
    "AT3G03350",
    "AT3G03360"
   ],
   "n_features": 1543,
   "default_features": [
    2634,
    5569,
    7708,
    5811,
    1564,
    7224,
    6329,
    479
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.633,
     "cut_percentile": 60
    },
    {
     "id": 5569,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.52,
     "cut_percentile": 40
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.518,
     "cut_percentile": 40
    },
    {
     "id": 5811,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.455,
     "cut_percentile": 40
    },
    {
     "id": 1564,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.604,
     "cut_percentile": 40
    },
    {
     "id": 7224,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.636,
     "cut_percentile": 60
    },
    {
     "id": 6329,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.573,
     "cut_percentile": 40
    },
    {
     "id": 479,
     "marks": "antisense:utr5",
     "nice": "antisense 5\u2032 UTR",
     "mcc_in_window": 0.869,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1545,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 4,
   "n_sites": 55,
   "bytes": 2565665,
   "selection": {
    "kind": "alternative",
    "headline": "an alternative splice site the canonical track drops",
    "why": "f586 reads 15.9 at chr3:793,413 on the sense strand, 38 bp from the canonical acceptor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads AG",
    "score": 15.9,
    "position": 5998,
    "evidence": {
     "verdict": "alternative_isoform",
     "coord": 793413,
     "dinuc": "AG",
     "shift_bp": 38,
     "in_full_annotation": true,
     "n_cases_in_window": 1
    }
   },
   "usage_check": {
    "verdict": "used",
    "headline": "used, and the site TAIR12 keeps",
    "summary": "AtRTD3 records 27 transcripts splicing at this acceptor and TAIR12 annotates it. The canonical acceptor 38 bp upstream, where f586 reads 0.00 against 15.92 here, is one TAIR12 drops.",
    "registers": [
     {
      "name": "Araport11 (Ensembl Plants 63)",
      "role": "transcript models",
      "result": "used",
      "detail": "AT3G03350 carries two transcripts. The acceptor is in the annotation, outside the canonical model the track draws."
     },
     {
      "name": "AtRTD3",
      "role": "long-read transcriptome",
      "result": "used",
      "detail": "27 transcripts splice at this acceptor."
     },
     {
      "name": "TAIR12",
      "role": "2026 reannotation",
      "result": "used",
      "detail": "Annotated in TAIR12. The canonical acceptor 38 bp upstream is not: TAIR12 drops it."
     }
    ]
   }
  },
  {
   "window_uid": "B00005450",
   "featured": false,
   "species": "athaliana",
   "species_binomial": "Arabidopsis thaliana",
   "family": "Brassicaceae",
   "order": "Brassicales",
   "assembly": "TAIR10",
   "chrom": "3",
   "start": 5412378,
   "end": 5420569,
   "strand": -1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "acceptor",
    "offset": 2961
   },
   "note": "an alternative splice site the canonical track drops \u2014 f586 reads 15.99 at chr3:5,415,301 on the sense strand, 16 bp from the canonical acceptor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads AG",
   "genes_sense": [
    "AT3G15980"
   ],
   "n_features": 1206,
   "default_features": [
    2634,
    5569,
    7708,
    7447,
    2798,
    7224,
    7944,
    3525
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.787,
     "cut_percentile": 40
    },
    {
     "id": 5569,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.639,
     "cut_percentile": 88
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.704,
     "cut_percentile": 40
    },
    {
     "id": 7447,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.515,
     "cut_percentile": 40
    },
    {
     "id": 2798,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.696,
     "cut_percentile": 40
    },
    {
     "id": 7224,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.864,
     "cut_percentile": 60
    },
    {
     "id": 7944,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.78,
     "cut_percentile": 40
    },
    {
     "id": 3525,
     "marks": "sense:utr5",
     "nice": "5\u2032 UTR",
     "mcc_in_window": 0.768,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1207,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 46,
   "bytes": 2264532,
   "selection": {
    "kind": "alternative",
    "headline": "an alternative splice site the canonical track drops",
    "why": "f586 reads 15.99 at chr3:5,415,301 on the sense strand, 16 bp from the canonical acceptor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads AG",
    "score": 15.99,
    "position": 5268,
    "evidence": {
     "verdict": "alternative_isoform",
     "coord": 5415301,
     "dinuc": "AG",
     "shift_bp": 16,
     "in_full_annotation": true,
     "n_cases_in_window": 2
    }
   },
   "usage_check": {
    "verdict": "used",
    "headline": "used, and the site TAIR12 keeps",
    "summary": "AtRTD3 records 24 transcripts splicing at this acceptor and TAIR12 annotates it. The canonical acceptor 16 bp downstream, where f586 reads 3.79 against 16.01 here, is one TAIR12 drops.",
    "registers": [
     {
      "name": "Araport11 (Ensembl Plants 63)",
      "role": "transcript models",
      "result": "used",
      "detail": "AT3G15980 carries seven transcripts. The acceptor is in the annotation, outside the canonical model the track draws."
     },
     {
      "name": "AtRTD3",
      "role": "long-read transcriptome",
      "result": "used",
      "detail": "24 transcripts splice at this acceptor."
     },
     {
      "name": "TAIR12",
      "role": "2026 reannotation",
      "result": "used",
      "detail": "Annotated in TAIR12. The canonical acceptor 16 bp downstream is not: TAIR12 drops it."
     }
    ]
   }
  },
  {
   "window_uid": "B00013816",
   "featured": false,
   "species": "osativa",
   "species_binomial": "Oryza sativa",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "IRGSP-1.0",
   "chrom": "5",
   "start": 14158564,
   "end": 14166755,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "acceptor",
    "offset": 5041
   },
   "note": "an alternative splice site the canonical track drops \u2014 f4131 reads 14.94 at chr5:14,163,098 on the sense strand, 174 bp from the canonical donor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads GT",
   "genes_sense": [
    "OsCatB"
   ],
   "n_features": 1878,
   "default_features": [
    2634,
    38,
    7708,
    962,
    7991,
    3686,
    1144,
    1068
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.656,
     "cut_percentile": 60
    },
    {
     "id": 38,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.645,
     "cut_percentile": 88
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.733,
     "cut_percentile": 40
    },
    {
     "id": 962,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.537,
     "cut_percentile": 40
    },
    {
     "id": 7991,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.619,
     "cut_percentile": 75
    },
    {
     "id": 3686,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.603,
     "cut_percentile": 40
    },
    {
     "id": 1144,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.534,
     "cut_percentile": 75
    },
    {
     "id": 1068,
     "marks": "sense:utr5",
     "nice": "5\u2032 UTR",
     "mcc_in_window": 0.72,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1880,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 2,
   "n_sites": 22,
   "bytes": 2497585,
   "selection": {
    "kind": "alternative",
    "headline": "an alternative splice site the canonical track drops",
    "why": "f4131 reads 14.94 at chr5:14,163,098 on the sense strand, 174 bp from the canonical donor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads GT",
    "score": 14.94,
    "position": 4533,
    "evidence": {
     "verdict": "alternative_isoform",
     "coord": 14163098,
     "dinuc": "GT",
     "shift_bp": 174,
     "in_full_annotation": true,
     "n_cases_in_window": 1
    }
   },
   "usage_check": {
    "verdict": "untested",
    "headline": "no independent register on this assembly",
    "summary": "The site is in the RAP-DB annotation, outside the canonical model the track draws. No long-read transcriptome or splicing atlas was consulted on IRGSP-1.0, so nothing independent bears on it.",
    "registers": [
     {
      "name": "RAP-DB via Ensembl Plants 59",
      "role": "transcript models",
      "result": "used",
      "detail": "The donor is annotated, outside the canonical model the track draws."
     },
     {
      "name": "None on IRGSP-1.0",
      "role": "long-read transcriptome, splicing atlas",
      "result": "not consulted",
      "detail": "No usage register for rice was consulted."
     }
    ]
   }
  },
  {
   "window_uid": "B00020628",
   "featured": false,
   "species": "zmays",
   "species_binomial": "Zea mays",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "Zm-B73-REFERENCE-NAM-5.0",
   "chrom": "9",
   "start": 29215287,
   "end": 29223478,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "donor",
    "offset": 3157
   },
   "note": "an alternative splice site the canonical track drops \u2014 f4131 reads 17.69 at chr9:29,223,261 on the sense strand, 162 bp from the canonical donor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads GT",
   "genes_sense": [
    "Zm00001eb379300",
    "Zm00001eb379310"
   ],
   "n_features": 1558,
   "default_features": [
    2634,
    2993,
    7708,
    2467,
    5215,
    8072,
    2425,
    5367
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.652,
     "cut_percentile": 40
    },
    {
     "id": 2993,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.619,
     "cut_percentile": 75
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.529,
     "cut_percentile": 40
    },
    {
     "id": 2467,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.466,
     "cut_percentile": 40
    },
    {
     "id": 5215,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.572,
     "cut_percentile": 40
    },
    {
     "id": 8072,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.543,
     "cut_percentile": 40
    },
    {
     "id": 2425,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.739,
     "cut_percentile": 88
    },
    {
     "id": 5367,
     "marks": "sense:utr3",
     "nice": "3\u2032 UTR",
     "mcc_in_window": 0.848,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1560,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 2,
   "n_sites": 20,
   "bytes": 2359359,
   "selection": {
    "kind": "alternative",
    "headline": "an alternative splice site the canonical track drops",
    "why": "f4131 reads 17.69 at chr9:29,223,261 on the sense strand, 162 bp from the canonical donor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads GT",
    "score": 17.69,
    "position": 7973,
    "evidence": {
     "verdict": "alternative_isoform",
     "coord": 29223261,
     "dinuc": "GT",
     "shift_bp": 162,
     "in_full_annotation": true,
     "n_cases_in_window": 1
    }
   },
   "usage_check": {
    "verdict": "untested",
    "headline": "no independent register on this assembly",
    "summary": "The site is in the Zm00001eb.1 annotation, outside the canonical model the track draws. No long-read transcriptome or splicing atlas was consulted on Zm-B73-REFERENCE-NAM-5.0, so nothing independent bears on it.",
    "registers": [
     {
      "name": "Zm00001eb.1 (Ensembl Plants 62)",
      "role": "transcript models",
      "result": "used",
      "detail": "The donor is annotated, outside the canonical model the track draws."
     },
     {
      "name": "None on Zm-B73-REFERENCE-NAM-5.0",
      "role": "long-read transcriptome, splicing atlas",
      "result": "not consulted",
      "detail": "No usage register for maize was consulted."
     }
    ]
   }
  },
  {
   "window_uid": "B00022102",
   "featured": false,
   "species": "zmays",
   "species_binomial": "Zea mays",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "Zm-B73-REFERENCE-NAM-5.0",
   "chrom": "9",
   "start": 77174247,
   "end": 77182438,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "acceptor",
    "offset": 4969
   },
   "note": "an alternative splice site the canonical track drops \u2014 f4131 reads 15.5 at chr9:77,180,079 on the sense strand, 164 bp from the canonical donor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads GT",
   "genes_sense": [
    "Zm00001eb384370"
   ],
   "n_features": 1463,
   "default_features": [
    2634,
    2993,
    7708,
    6428,
    5554,
    49,
    4553,
    5782
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.727,
     "cut_percentile": 60
    },
    {
     "id": 2993,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.666,
     "cut_percentile": 60
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.672,
     "cut_percentile": 40
    },
    {
     "id": 6428,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.493,
     "cut_percentile": 40
    },
    {
     "id": 5554,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.709,
     "cut_percentile": 40
    },
    {
     "id": 49,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.745,
     "cut_percentile": 40
    },
    {
     "id": 4553,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.792,
     "cut_percentile": 40
    },
    {
     "id": 5782,
     "marks": "sense:utr3",
     "nice": "3\u2032 UTR",
     "mcc_in_window": 0.671,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1464,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 17,
   "bytes": 2479083,
   "selection": {
    "kind": "alternative",
    "headline": "an alternative splice site the canonical track drops",
    "why": "f4131 reads 15.5 at chr9:77,180,079 on the sense strand, 164 bp from the canonical donor. The canonical transcript does not use that site, but another annotated isoform does, and the base reads GT",
    "score": 15.5,
    "position": 5831,
    "evidence": {
     "verdict": "alternative_isoform",
     "coord": 77180079,
     "dinuc": "GT",
     "shift_bp": 164,
     "in_full_annotation": true,
     "n_cases_in_window": 3
    }
   },
   "usage_check": {
    "verdict": "untested",
    "headline": "no independent register on this assembly",
    "summary": "The site is in the Zm00001eb.1 annotation, outside the canonical model the track draws. No long-read transcriptome or splicing atlas was consulted on Zm-B73-REFERENCE-NAM-5.0, so nothing independent bears on it.",
    "registers": [
     {
      "name": "Zm00001eb.1 (Ensembl Plants 62)",
      "role": "transcript models",
      "result": "used",
      "detail": "The donor is annotated, outside the canonical model the track draws."
     },
     {
      "name": "None on Zm-B73-REFERENCE-NAM-5.0",
      "role": "long-read transcriptome, splicing atlas",
      "result": "not consulted",
      "detail": "No usage register for maize was consulted."
     }
    ]
   }
  },
  {
   "window_uid": "B00004408",
   "featured": false,
   "species": "athaliana",
   "species_binomial": "Arabidopsis thaliana",
   "family": "Brassicaceae",
   "order": "Brassicales",
   "assembly": "TAIR10",
   "chrom": "1",
   "start": 21733275,
   "end": 21741466,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "donor",
    "offset": 4469
   },
   "note": "a splice site no annotated transcript uses \u2014 f586 reads 11.24 at chr1:21,735,155 on the sense strand, reading AG, inside the annotated intron whose own acceptor is 42 bp downstream. No transcript in Araport11, AtRTD3 or TAIR12 splices there, and PastDB measures the site in use",
   "genes_sense": [
    "AT1G58525"
   ],
   "n_features": 1372,
   "default_features": [
    2634,
    2993,
    7708,
    292,
    5342,
    258,
    4062,
    7056
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.769,
     "cut_percentile": 40
    },
    {
     "id": 2993,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.645,
     "cut_percentile": 40
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.812,
     "cut_percentile": 40
    },
    {
     "id": 292,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.492,
     "cut_percentile": 40
    },
    {
     "id": 5342,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.755,
     "cut_percentile": 40
    },
    {
     "id": 258,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.735,
     "cut_percentile": 40
    },
    {
     "id": 4062,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.886,
     "cut_percentile": 40
    },
    {
     "id": 7056,
     "marks": "sense:utr5",
     "nice": "5\u2032 UTR",
     "mcc_in_window": 0.956,
     "cut_percentile": 60
    }
   ],
   "coverage": {
    "features_stored": 1373,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 24,
   "bytes": 2280212,
   "selection": {
    "kind": "unannotated",
    "headline": "a splice site no annotated transcript uses",
    "why": "f586 reads 11.24 at chr1:21,735,155 on the sense strand, reading AG, inside the annotated intron whose own acceptor is 42 bp downstream. No transcript in Araport11, AtRTD3 or TAIR12 splices there, and PastDB measures the site in use",
    "score": 11.24,
    "position": 1879,
    "evidence": {
     "verdict": "unannotated_but_used",
     "coord": 21735155,
     "dinuc": "AG",
     "nearest_full_bp": 42,
     "in_full_annotation": false,
     "n_cases_in_window": 1,
     "nearest_site_activation": 14.78
    }
   },
   "usage_check": {
    "verdict": "used",
    "headline": "a real minor acceptor, below what a transcript model carries",
    "summary": "PastDB catalogues this acceptor as an alternative 3\u2032 splice site of the same intron, used at up to 7.5% and 0.11% on average across its 202 RNA-seq samples, against 99.8% for the dominant site 42 bp downstream. No transcript-level annotation draws a model through usage that low, and none of the three does.",
    "registers": [
     {
      "name": "Araport11 (Ensembl Plants 63)",
      "role": "transcript models",
      "result": "absent",
      "detail": "Seven transcripts of AT1G58520. None splices here."
     },
     {
      "name": "AtRTD3",
      "role": "long-read transcriptome",
      "result": "absent",
      "detail": "Five models of AT1G58520, two of them Iso-seq, plus two read-through models. None splices here; six use the dominant acceptor 42 bp downstream."
     },
     {
      "name": "TAIR12",
      "role": "2026 reannotation",
      "result": "absent",
      "detail": "No site here. The dominant acceptor 42 bp downstream is unchanged."
     },
     {
      "name": "PastDB",
      "role": "splicing atlas, 202 RNA-seq samples",
      "result": "used",
      "detail": "Alternative 3\u2032 splice site AthALTA0041238-3/3: average usage 0.11%, up to 7.5%."
     }
    ]
   }
  },
  {
   "window_uid": "B00006020",
   "featured": false,
   "species": "athaliana",
   "species_binomial": "Arabidopsis thaliana",
   "family": "Brassicaceae",
   "order": "Brassicales",
   "assembly": "TAIR10",
   "chrom": "5",
   "start": 5419612,
   "end": 5427803,
   "strand": -1,
   "n_bases": 8191,
   "split": "final_test",
   "anchor": {
    "kind": "acceptor",
    "offset": 5443
   },
   "note": "a splice site no annotated transcript uses \u2014 f586 reads 10.62 at chr5:5,427,495 on the sense strand, reading AG, 23 bp upstream of BGLU2\u2019s start codon in sequence the annotation calls intergenic, because the gene model begins at that codon and carries no 5\u2032 UTR. No transcript in Araport11, AtRTD3 or TAIR12 splices there, and BGLU2 is not expressed in PastDB, so the site cannot be tested either way",
   "genes_sense": [
    "GLN1",
    "BGLU2"
   ],
   "n_features": 1252,
   "default_features": [
    2634,
    586,
    7708,
    2467,
    4037,
    7224,
    4928,
    3525
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.713,
     "cut_percentile": 60
    },
    {
     "id": 586,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.636,
     "cut_percentile": 60
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.8,
     "cut_percentile": 40
    },
    {
     "id": 2467,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.584,
     "cut_percentile": 40
    },
    {
     "id": 4037,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.631,
     "cut_percentile": 40
    },
    {
     "id": 7224,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.706,
     "cut_percentile": 60
    },
    {
     "id": 4928,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.781,
     "cut_percentile": 60
    },
    {
     "id": 3525,
     "marks": "sense:utr5",
     "nice": "5\u2032 UTR",
     "mcc_in_window": 0.645,
     "cut_percentile": 60
    }
   ],
   "coverage": {
    "features_stored": 1253,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 2,
   "n_sites": 43,
   "bytes": 2529968,
   "selection": {
    "kind": "unannotated",
    "headline": "a splice site no annotated transcript uses",
    "why": "f586 reads 10.62 at chr5:5,427,495 on the sense strand, reading AG, 23 bp upstream of BGLU2\u2019s start codon in sequence the annotation calls intergenic, because the gene model begins at that codon and carries no 5\u2032 UTR. No transcript in Araport11, AtRTD3 or TAIR12 splices there, and BGLU2 is not expressed in PastDB, so the site cannot be tested either way",
    "score": 10.62,
    "position": 308,
    "evidence": {
     "verdict": "unannotated_untestable",
     "coord": 5427495,
     "dinuc": "AG",
     "nearest_full_bp": 197,
     "in_full_annotation": false,
     "n_cases_in_window": 1,
     "nearest_site_activation": 0.0
    }
   },
   "usage_check": {
    "verdict": "untestable",
    "headline": "the gene is not expressed, so nothing can be measured",
    "summary": "BGLU2 has one transcript in every register and no annotated 5\u2032 UTR, so no alternative isoform was ever available to hold the site, and its median expression across PastDB\u2019s 40 abiotic-stress samples is 0.00 cRPKM, so no read can confirm or refute it. f586 also reads 0.00 at BGLU2\u2019s own first annotated acceptor 197 bp away.",
    "registers": [
     {
      "name": "Araport11 (Ensembl Plants 63)",
      "role": "transcript models",
      "result": "absent",
      "detail": "AT5G16580 has one transcript and no annotated 5\u2032 UTR: the model begins at the start codon 23 bp downstream of the call."
     },
     {
      "name": "AtRTD3",
      "role": "long-read transcriptome",
      "result": "no data",
      "detail": "One model for AT5G16580, inherited from AtRTD2 rather than Iso-seq. The neighbouring GLN1 carries eight models, seven of them Iso-seq, so the locus was sampled and this gene was not resolved."
     },
     {
      "name": "TAIR12",
      "role": "2026 reannotation",
      "result": "absent",
      "detail": "No site here. BGLU2\u2019s own acceptor and donor are unchanged."
     },
     {
      "name": "PastDB",
      "role": "splicing atlas, 202 RNA-seq samples",
      "result": "no data",
      "detail": "No alternative 3\u2032 splice site event for AT5G16580, and a median expression of 0.00 cRPKM over 40 samples."
     }
    ]
   }
  },
  {
   "window_uid": "B00020610",
   "featured": false,
   "species": "zmays",
   "species_binomial": "Zea mays",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "Zm-B73-REFERENCE-NAM-5.0",
   "chrom": "9",
   "start": 156445609,
   "end": 156453800,
   "strand": -1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "donor",
    "offset": 4244
   },
   "note": "a splice site no annotated transcript uses \u2014 f586 reads 13.05 at chr9:156,453,517 on the sense strand, reading AG, inside an annotated 3\u2032 UTR whose nearest annotated acceptor is 193 bp away and reads 12.88. No transcript in Zm00001eb.1 splices there, and no long-read transcriptome or splicing atlas was consulted on this assembly, so the site is untested",
   "genes_sense": [
    "Zm00001eb401630"
   ],
   "n_features": 1481,
   "default_features": [
    2634,
    2993,
    7708
   ],
   "lane_rationale": [
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.625,
     "cut_percentile": 40
    },
    {
     "id": 2993,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.526,
     "cut_percentile": 60
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.614,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1483,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 17,
   "bytes": 2419041,
   "selection": {
    "kind": "unannotated",
    "headline": "a splice site no annotated transcript uses",
    "why": "f586 reads 13.05 at chr9:156,453,517 on the sense strand, reading AG, inside an annotated 3\u2032 UTR whose nearest annotated acceptor is 193 bp away and reads 12.88. No transcript in Zm00001eb.1 splices there, and no long-read transcriptome or splicing atlas was consulted on this assembly, so the site is untested",
    "score": 13.05,
    "position": 283,
    "evidence": {
     "verdict": "unannotated_untested",
     "coord": 156453517,
     "dinuc": "AG",
     "nearest_full_bp": 193,
     "in_full_annotation": false,
     "n_cases_in_window": 1,
     "nearest_site_activation": 12.88
    }
   },
   "usage_check": {
    "verdict": "untested",
    "headline": "no register on this assembly can test it",
    "summary": "Zm00001eb.1 is the only annotation released on this assembly, and no long-read transcriptome or splicing atlas was consulted, so the site is untested rather than contradicted. f586 reads 12.88 at the nearest annotated acceptor 193 bp away, against 13.05 here.",
    "registers": [
     {
      "name": "Zm00001eb.1 (Ensembl Plants 62)",
      "role": "transcript models",
      "result": "absent",
      "detail": "No transcript splices here."
     },
     {
      "name": "None on Zm-B73-REFERENCE-NAM-5.0",
      "role": "long-read transcriptome, splicing atlas",
      "result": "not consulted",
      "detail": "No usage register for maize was consulted."
     }
    ]
   }
  },
  {
   "window_uid": "B00005282",
   "featured": false,
   "species": "athaliana",
   "species_binomial": "Arabidopsis thaliana",
   "family": "Brassicaceae",
   "order": "Brassicales",
   "assembly": "TAIR10",
   "chrom": "5",
   "start": 19483610,
   "end": 19491801,
   "strand": 1,
   "n_bases": 8191,
   "split": "final_test",
   "anchor": {
    "kind": "acceptor",
    "offset": 1440
   },
   "note": "the annotation itself moved \u2014 TAIR12 moves the acceptor of this ELP1 intron 22 bp, from chr5:19,486,348 to chr5:19,486,370 in the TAIR10 frame. TAIR12's own coordinate is 22,640,459 on the Col-CC v2 assembly; no TAIR10-to-TAIR12 liftover is published, so the two frames were tied together by anchoring on sequence identical in both. The exon count is unchanged at 18, and 33 other splice boundaries in this window are untouched. A whole-genome alignment of the two assemblies, the route the usage register takes, returns the same two coordinates: it drops the acceptor at 19,486,348 and gains one at 19,486,370, whose TAIR12 coordinate is 22,640,459",
   "genes_sense": [
    "ELP1"
   ],
   "n_features": 1438,
   "default_features": [
    586,
    2634,
    38,
    4902,
    7447,
    4037,
    49,
    4553,
    0
   ],
   "lane_rationale": [
    {
     "id": 586,
     "marks": "the claim this window is here for",
     "nice": "the contested site",
     "mcc_in_window": null,
     "cut_percentile": null
    },
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.703,
     "cut_percentile": 60
    },
    {
     "id": 38,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.581,
     "cut_percentile": 75
    },
    {
     "id": 4902,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.835,
     "cut_percentile": 40
    },
    {
     "id": 7447,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.536,
     "cut_percentile": 40
    },
    {
     "id": 4037,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.597,
     "cut_percentile": 40
    },
    {
     "id": 49,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.615,
     "cut_percentile": 40
    },
    {
     "id": 4553,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.717,
     "cut_percentile": 40
    },
    {
     "id": 0,
     "marks": "sense:utr3",
     "nice": "3\u2032 UTR",
     "mcc_in_window": 0.827,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1440,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 2,
   "n_sites": 44,
   "bytes": 2402418,
   "selection": {
    "kind": "changed",
    "headline": "the annotation itself moved",
    "why": "TAIR12 moves the acceptor of this ELP1 intron 22 bp, from chr5:19,486,348 to chr5:19,486,370 in the TAIR10 frame. TAIR12's own coordinate is 22,640,459 on the Col-CC v2 assembly; no TAIR10-to-TAIR12 liftover is published, so the two frames were tied together by anchoring on sequence identical in both. The exon count is unchanged at 18, and 33 other splice boundaries in this window are untouched. A whole-genome alignment of the two assemblies, the route the usage register takes, returns the same two coordinates: it drops the acceptor at 19,486,348 and gains one at 19,486,370, whose TAIR12 coordinate is 22,640,459",
    "score": 22,
    "position": 2737,
    "evidence": {
     "verdict": "annotation_changed",
     "gene_id": "AT5G48090",
     "gene_name": "ELP1",
     "event": "moved_acceptor",
     "old_coord": 19486348,
     "new_coord": 19486370,
     "shift_bp": 22,
     "window_index_old": 2737,
     "window_index_new": 2759,
     "native_coord_tair12": 22640459,
     "local_offset": 3154089,
     "sequence_anchored": true,
     "exons_old": 18,
     "exons_new": 18,
     "changed_boundaries_in_window": 2,
     "unchanged_boundaries_in_window": 33,
     "isoform_repick": false,
     "edge_distance_bp": 2737,
     "old_release": "Araport11 (Ensembl Plants 59)",
     "new_release": "TAIR12_1Feb26",
     "new_release_assembly": "Col-CC v2 (GCA_978657495.1)"
    }
   }
  },
  {
   "window_uid": "B00014342",
   "featured": false,
   "species": "osativa",
   "species_binomial": "Oryza sativa",
   "family": "Poaceae",
   "order": "Poales",
   "assembly": "IRGSP-1.0",
   "chrom": "7",
   "start": 5015096,
   "end": 5023287,
   "strand": 1,
   "n_bases": 8191,
   "split": "interp_train",
   "anchor": {
    "kind": "acceptor",
    "offset": 3164
   },
   "note": "the annotation itself moved \u2014 RAP-DB's 2026-02-05 release moves this donor 26 bp, from chr7:5,018,090 to chr7:5,018,116, on the same IRGSP-1.0 assembly the window is cut from, so the shift needs no re-mapping. The exon count is unchanged at 9, and 13 other splice boundaries in this window are untouched",
   "genes_sense": [
    "Os07g0192900"
   ],
   "n_features": 1343,
   "default_features": [
    4131,
    2634,
    586,
    7708,
    5811,
    1564,
    3795,
    734,
    643
   ],
   "lane_rationale": [
    {
     "id": 4131,
     "marks": "the claim this window is here for",
     "nice": "the contested site",
     "mcc_in_window": null,
     "cut_percentile": null
    },
    {
     "id": 2634,
     "marks": "donor",
     "nice": "splice donor",
     "mcc_in_window": 0.667,
     "cut_percentile": 60
    },
    {
     "id": 586,
     "marks": "acceptor",
     "nice": "splice acceptor",
     "mcc_in_window": 0.615,
     "cut_percentile": 75
    },
    {
     "id": 7708,
     "marks": "intron",
     "nice": "intron",
     "mcc_in_window": 0.745,
     "cut_percentile": 60
    },
    {
     "id": 5811,
     "marks": "cds",
     "nice": "coding sequence",
     "mcc_in_window": 0.588,
     "cut_percentile": 40
    },
    {
     "id": 1564,
     "marks": "codon position 1",
     "nice": "codon position 1",
     "mcc_in_window": 0.695,
     "cut_percentile": 40
    },
    {
     "id": 3795,
     "marks": "codon position 2",
     "nice": "codon position 2",
     "mcc_in_window": 0.594,
     "cut_percentile": 60
    },
    {
     "id": 734,
     "marks": "codon position 3",
     "nice": "codon position 3",
     "mcc_in_window": 0.663,
     "cut_percentile": 40
    },
    {
     "id": 643,
     "marks": "sense:utr3",
     "nice": "3\u2032 UTR",
     "mcc_in_window": 0.641,
     "cut_percentile": 40
    }
   ],
   "coverage": {
    "features_stored": 1344,
    "panel_depth": 8,
    "rule": "every feature reaching the top 8 at any base of this window, plus the named features; each carries a complete stride-1 track",
    "note": "This dictionary is BatchTopK with k=64, so at any base the 64 strongest features are non-zero and every other feature is exactly zero. A stored track is therefore complete over the whole window: where it reads zero, the feature did not fire. A feature not stored here never reaches the top 8 at any base, so it can never appear in the panel either."
   },
   "lane_selection": {
    "rule": "one lane per genomic category, each the feature that matches that category best in this window, searched over every feature the window stores",
    "score": "Matthews correlation against the annotated category",
    "caveat": "the threshold is the best of six percentiles of the feature's own values in this window, so the score describes this picture and is not an out-of-sample estimate",
    "phase_source": "codon positions are derived from the canonical transcripts' CDS in transcription order; the label tracks do not carry reading frame"
   },
   "n_genes": 1,
   "n_sites": 17,
   "bytes": 2436810,
   "selection": {
    "kind": "changed",
    "headline": "the annotation itself moved",
    "why": "RAP-DB's 2026-02-05 release moves this donor 26 bp, from chr7:5,018,090 to chr7:5,018,116, on the same IRGSP-1.0 assembly the window is cut from, so the shift needs no re-mapping. The exon count is unchanged at 9, and 13 other splice boundaries in this window are untouched",
    "score": 26,
    "position": 2993,
    "evidence": {
     "verdict": "annotation_changed",
     "gene_id": "Os07g0192900",
     "event": "moved_donor",
     "old_coord": 5018090,
     "new_coord": 5018116,
     "shift_bp": 26,
     "window_index_old": 2993,
     "window_index_new": 3019,
     "exons_old": 9,
     "exons_new": 9,
     "changed_boundaries_in_window": 3,
     "unchanged_boundaries_in_window": 13,
     "isoform_repick": false,
     "edge_distance_bp": 2993,
     "old_release": "RAP-DB via Ensembl Plants 59 (genebuild 2022-09-RAPDB)",
     "new_release": "RAP-DB IRGSP-1.0 representative set, 2026-02-05"
    }
   }
  }
 ],
 "sources": {
  "intro": "Every window carries two gene registers on one assembly. The first is the annotation the model's features were scored against, and it is frozen: Araport11 for Arabidopsis and rice's RAP-DB build via Ensembl Plants 59, Zm00001eb.1 for maize and ITAG3.0 for tomato via Ensembl Plants 62. The second holds the most recent gene models published on that same assembly, so annotation that has moved since is visible without any coordinate being re-mapped. For rice and tomato that is a genuinely newer release; for Arabidopsis and maize the newest file restates the same gene models.",
  "species": [
   {
    "binomial": "Arabidopsis thaliana",
    "code": "athaliana",
    "assembly": "TAIR10",
    "assembly_accession": "GCA_000001735.1",
    "annotation_build": "Araport11 (Ensembl Plants 59)",
    "annotation_current": "Araport11 (Ensembl Plants 63)",
    "released": "",
    "url": "https://ftp.ebi.ac.uk/ensemblgenomes/pub/release-63/plants/gtf/arabidopsis_thaliana/Arabidopsis_thaliana.TAIR10.63.gtf.gz",
    "citation": "Cheng C-Y, Krishnakumar V, Chan AP, Thibaud-Nissen F, Schobel S, Town CD. Araport11: a complete reannotation of the Arabidopsis thaliana reference genome. The Plant Journal 89:789-804 (2017). doi:10.1111/tpj.13415",
    "licence": "Ensembl Plants places no reuse restrictions on its annotation files.",
    "currency": "current",
    "currency_note": "Ensembl Plants 63 still carries the Araport11 genebuild on TAIR10 and is coordinate-identical to release 59: across all 888,095 GTF feature rows there is not one differing chromosome, type, start, end, strand, gene or transcript id. Ensembl Plants records this genebuild as \"Araport11\" with no release date, so no release date is given here; the models are those of Cheng et al. (2017).",
    "superseded_by": "TAIR12 on Col-CC v2 (GCA_978657495.1, annotation TAIR12_1Feb26, released 2026-02-13) is a new assembly of 142,481,245 bp against TAIR10's 119,667,750 bp. No liftover chain has been published and the per-locus offsets are not global, so no TAIR12 gene model is drawn in the annotation register. Individual sites are still asked of it: the usage register below looks each one up through a whole-genome alignment of the two assemblies."
   },
   {
    "binomial": "Oryza sativa Japonica Group",
    "code": "osativa",
    "assembly": "IRGSP-1.0",
    "assembly_accession": "GCA_001433935.1",
    "annotation_build": "RAP-DB via Ensembl Plants 59 (genebuild 2022-09-RAPDB)",
    "annotation_current": "RAP-DB IRGSP-1.0 representative set, 2026-02-05",
    "released": "2026-02-05",
    "url": "https://rapdb.dna.naro.go.jp/download/archive/irgsp1/IRGSP-1.0_representative_2026-02-05.tar.gz",
    "citation": "Sakai H, Lee SS, Tanaka T, et al. Rice Annotation Project Database (RAP-DB): An Integrative and Interactive Database for Rice Genomics. Plant and Cell Physiology 54:e6 (2013). doi:10.1093/pcp/pcs183; Kawahara Y, de la Bastide M, Hamilton JP, et al. Improvement of the Oryza sativa Nipponbare reference genome using next generation sequence and optical map data. Rice 6:4 (2013). doi:10.1186/1939-8433-6-4",
    "licence": "RAP-DB content is (c) NARO. No explicit reuse grant is published, so the register here is a coordinate extract, not a redistribution of the release.",
    "currency": "upgraded",
    "currency_note": "The 2026-02-05 representative set is on the same IRGSP-1.0 assembly and retires 1,995 of the 37,864 RAP loci the build carried while adding 192, for 36,061 loci. Across the five rice windows here every build locus survives: Os02g0291000 (OsCBL7) gains a 5' exon and 832 bp of span, and two loci pick up updated gene symbols.",
    "superseded_by": "NCBI has moved its rice reference to AGIS1.0 (GCF_034140825.1), a different assembly that cannot be overlaid on IRGSP-1.0 windows."
   },
   {
    "binomial": "Zea mays",
    "code": "zmays",
    "assembly": "Zm-B73-REFERENCE-NAM-5.0",
    "assembly_accession": "GCA_902167145.1",
    "annotation_build": "Zm00001eb.1 (Ensembl Plants 62)",
    "annotation_current": "Zm00001eb.1 (Ensembl Plants 63)",
    "released": "2019-12",
    "url": "https://ftp.ebi.ac.uk/ensemblgenomes/pub/release-63/plants/gtf/zea_mays/Zea_mays.Zm-B73-REFERENCE-NAM-5.0.63.gtf.gz",
    "citation": "Hufford MB, Seetharam AS, Woodhouse MR, et al. De novo assembly, annotation, and comparative analysis of 26 diverse maize genomes. Science 373:655-662 (2021). doi:10.1126/science.abg5289",
    "licence": "Ensembl Plants places no reuse restrictions on its annotation files.",
    "currency": "current",
    "currency_note": "Zm00001eb.1 is the only annotation ever released on this assembly, and Ensembl Plants 62, Ensembl Plants 63 and MaizeGDB carry the same models: across all 1,302,218 GTF feature rows releases 62 and 63 differ in nothing but transcript naming metadata. The release date shown is the genebuild date Ensembl records for it (2019-12-CSHL).",
    "superseded_by": ""
   },
   {
    "binomial": "Solanum lycopersicum",
    "code": "slycopersicum",
    "assembly": "SL3.0",
    "assembly_accession": "GCA_000188115.3",
    "annotation_build": "ITAG3.0 via Ensembl Plants 62 (genebuild 2017-02-SOL)",
    "annotation_current": "ITAG3.2",
    "released": "2017-06-15",
    "url": "https://solgenomics.net/ftp/tomato_genome/annotation/ITAG3.2_release/ITAG3.2_gene_models.gff",
    "citation": "The Tomato Genome Consortium. The tomato genome sequence provides insights into fleshy fruit evolution. Nature 485:635-641 (2012). doi:10.1038/nature11119; Fernandez-Pozo N, Menda N, Edwards JD, et al. The Sol Genomics Network (SGN)-from genotype to phenotype to breeding. Nucleic Acids Research 43:D1036-D1041 (2014). doi:10.1093/nar/gku1195",
    "licence": "SGN publishes no licence with ITAG3.2. Where redistribution matters, prefer the public Ensembl Plants 62 SL3.0 extract; the register here is a coordinate extract only.",
    "currency": "upgraded",
    "currency_note": "ITAG3.2 is on the same SL3.0 assembly - all 12 chromosome lengths are byte-identical - and holds 35,768 loci against 34,761 in ITAG3.0: 1,773 retired, 2,780 added and 1,036 with a changed span. None of that reaches these five windows: all ten loci in them are identical between ITAG3.0 and ITAG3.2. ITAG3.2 carries one mRNA per gene and no non-coding genes at all, so non-coding loci are carried over from the build-time Ensembl Plants 62 SL3.0 file rather than dropped.",
    "superseded_by": "SL4.0 with ITAG4.1 (2020-01-28) and SL5.0 (Zhou Y et al., Nature 606:527-534, 2022) both supersede SL3.0. No SL3.0 to SL4.0 chain has been published and the coordinate shift is non-linear, so neither can be overlaid on these windows. Ensembl Plants dropped SL3.0 after release 62 and carries tomato only on SL4.0 (GCA_000188115.5) from release 63."
   }
  ],
  "caption": "Gene annotation behind each window: the release the feature tracks were scored against, and the most current release available on the same assembly. Newer assemblies are named but never overlaid - no liftover chain is published for any of them.",
  "notes": [
   "Arabidopsis: TAIR12 on the Col-CC v2 assembly (INSDC GCA_978657495.1, annotation TAIR12_1Feb26, released 2026-02-13) is the newest Arabidopsis annotation, but it belongs to a different assembly - 142,481,245 bp against TAIR10's 119,667,750 bp. No liftover chain has been published anywhere and the per-locus offsets are not a global shift, so no TAIR12 gene model is placed on these TAIR10 windows and none is drawn in the annotation register. Single sites are a different question, and the usage register answers it by aligning the two genomes and asking whether TAIR12 annotates the one base in question.",
   "Tomato: SL4.0 (ITAG4.1, released 2020-01-28) and SL5.0 (Zhou Y et al., Nature 606:527-534, 2022) both supersede SL3.0, and Ensembl Plants stopped carrying SL3.0 after release 62. No SL3.0 to SL4.0 chain is published and the coordinate shift is non-linear, so the current register stays on SL3.0 with ITAG3.2, the last annotation released on that assembly.",
   "Rice: NCBI has moved its rice reference to AGIS1.0 (GCF_034140825.1). That is a different assembly, so the current register stays on IRGSP-1.0 with the RAP-DB representative set of 2026-02-05. RAP-DB has also moved host: use rapdb.dna.naro.go.jp, as the older rapdb.dna.affrc.go.jp host no longer responds.",
   "Maize: no B73 v6 assembly exists and Zm00001eb.1 is the only annotation released on Zm-B73-REFERENCE-NAM-5.0, so the build and current registers are the same gene models read from a newer file.",
   "Coordinates were never re-mapped. Both registers are window-relative base indices into the 8,191 stored bases, with 'to' inclusive and clipped to the window, and both split by orientation the same way: 'sense' means the gene's genomic strand matches the window's strand. The mapping was verified by refetching each window's reference sequence and matching its sha256 against the stored seq_sha256, and by reproducing 50 of the 51 build-register gene placements exactly from the current files. The one that moved, Os02g0291000, moved because RAP-DB extended the gene.",
   "Every source file was checksummed before parsing. The sha256 recorded in each window's genes_current is the sha256 of the uncompressed file that was parsed, the same convention already used by provenance.annotation_build_sha256, except for rice where it is the sha256 of the distributed archive whose three member GFF files were parsed together.",
   "Non-coding genes in ITAG3.2: the release contains 35,768 loci and every one of them is protein-coding, against 1,396 non-coding loci in the Ensembl Plants 62 SL3.0 build. The tomato current register is therefore ITAG3.2 plus any non-coding locus from the Ensembl Plants 62 file that ITAG3.2 does not carry, so switching register cannot silently delete a non-coding gene. Across the five tomato windows here that carry-over adds nothing, because none of them holds a non-coding gene."
  ],
  "citations": [
   "Cheng C-Y, Krishnakumar V, Chan AP, Thibaud-Nissen F, Schobel S, Town CD. Araport11: a complete reannotation of the Arabidopsis thaliana reference genome. The Plant Journal 89:789-804 (2017). doi:10.1111/tpj.13415",
   "Reiser L, Proia A, Bakker E, et al. Recent major changes to TAIR: updates to the database, website, and Arabidopsis genome. GENETICS 232(4) (2026). doi:10.1093/genetics/iyaf248",
   "Sakai H, Lee SS, Tanaka T, et al. Rice Annotation Project Database (RAP-DB): An Integrative and Interactive Database for Rice Genomics. Plant and Cell Physiology 54:e6 (2013). doi:10.1093/pcp/pcs183",
   "Kawahara Y, de la Bastide M, Hamilton JP, et al. Improvement of the Oryza sativa Nipponbare reference genome using next generation sequence and optical map data. Rice 6:4 (2013). doi:10.1186/1939-8433-6-4",
   "Hufford MB, Seetharam AS, Woodhouse MR, et al. De novo assembly, annotation, and comparative analysis of 26 diverse maize genomes. Science 373:655-662 (2021). doi:10.1126/science.abg5289",
   "The Tomato Genome Consortium. The tomato genome sequence provides insights into fleshy fruit evolution. Nature 485:635-641 (2012). doi:10.1038/nature11119",
   "Fernandez-Pozo N, Menda N, Edwards JD, et al. The Sol Genomics Network (SGN)-from genotype to phenotype to breeding. Nucleic Acids Research 43:D1036-D1041 (2014). doi:10.1093/nar/gku1195",
   "Zhou Y, Zhang Z, Bao Z, et al. Graph pangenome captures missing heritability and empowers tomato breeding. Nature 606:527-534 (2022). doi:10.1038/s41586-022-04808-9"
  ],
  "usage_registers": {
   "intro": "The annotations above say which splice sites a released gene model carries. They cannot say whether a site is used, only whether someone drew a transcript through it. Every window where a feature marks a site the annotation does not carry was therefore put to three further registers, all of them on TAIR10, so no coordinate is re-mapped.",
   "caption": "The three usage registers, all on the TAIR10 assembly the Arabidopsis windows are cut from.",
   "rows": [
    {
     "name": "AtRTD3",
     "role": "long-read transcriptome",
     "version": "GTF 21Feb22, TranSuite transfix build",
     "scope": "Arabidopsis thaliana, TAIR10",
     "content": "169,503 transcripts over 40,932 genes with 183,568 splice junctions, assembled primarily from PacBio Iso-seq reads.",
     "url": "https://ics.hutton.ac.uk/atRTD/RTD3/",
     "citation": "Zhang R, Kuo R, Coulter M, Calixto CPG, Entizne JC, Guo W, et al. A high-resolution single-molecule sequencing-based Arabidopsis transcriptome using novel methods of Iso-seq analysis. Genome Biology 23:149 (2022). doi:10.1186/s13059-022-02711-0",
     "caveat": "12,506 of its transcripts are inherited from Araport11 and 24,809 from AtRTD2, so a model in AtRTD3 is not by itself long-read evidence. The origin tag is read before a transcript is counted as Iso-seq support."
    },
    {
     "name": "TAIR12",
     "role": "2026 reannotation",
     "version": "TAIR12_1Feb26 on Col-CC v2, GCA_978657495.1",
     "scope": "Arabidopsis thaliana, brought into the TAIR10 frame",
     "content": "The first Arabidopsis reannotation since Araport11, on the telomere-to-telomere Col-CC assembly.",
     "url": "https://www.ebi.ac.uk/ena/browser/view/PRJEB100887",
     "citation": "Reiser et al. (2026), TAIR12. BioProject PRJEB100887.",
     "caveat": "TAIR12 is built on a different assembly, so its sites were carried into TAIR10 coordinates through a whole-genome alignment rather than a published liftover chain. Its models were rebuilt from the GenBank flatfile because the distributed GFF3 is lossy for isoforms."
    },
    {
     "name": "PastDB",
     "role": "splicing atlas",
     "version": "araTha10, vast-tools v2.5.1",
     "scope": "Arabidopsis thaliana, TAIR10",
     "content": "Alternative-splicing usage across 202 RNA-seq samples spanning tissues, developmental stages and stress conditions, reported as PSI for exons and PSU for alternative splice sites.",
     "url": "https://pastdb.crg.eu",
     "citation": "Mart\u00edn G, M\u00e1rquez Y, Duque P, Irimia M. Alternative splicing landscapes in Arabidopsis thaliana across tissues and stress conditions highlight major functional differences with animals. Genome Biology 22:35 (2021). doi:10.1186/s13059-020-02258-y",
     "caveat": "PastDB quantifies the splice sites its own event library defines. A site the library does not carry has no usage number, which is not the same as a usage of zero, and a gene that is not expressed has no measurable usage at all."
    }
   ],
   "note": "No equivalent long-read transcriptome or splicing atlas was consulted for rice, tomato or maize. The disagreements in those genomes are recorded as untested, not as confirmed or refuted.",
   "verdicts": [
    {
     "key": "used",
     "label": "Used",
     "gloss": "an independent register records the site in use"
    },
    {
     "key": "untestable",
     "label": "Untestable",
     "gloss": "the registers cover the locus but no measurement is possible there"
    },
    {
     "key": "untested",
     "label": "Untested",
     "gloss": "no usage register was consulted on this assembly"
    }
   ]
  }
 }
}