Expand dataset documentation

README.md

@@ -1,38 +1,149 @@
 ---
 pretty_name: "GAMBA Functional Region Multiclass"
 license: mit
-tags: [biology, genomics, genome-language-model, benchmark, hg38, parquet]
+tags:
+- biology
+- genomics
+- dna
+- regulatory-genomics
+- genome-language-model
+- multiclass-classification
+- benchmark
+- hg38
+- parquet
 configs:
 - config_name: full-causal
-  data_files: [{split: all, path: functional-multiclass-gamba-full-causal.parquet}]
+  data_files:
+  - split: all
+    path: functional-multiclass-gamba-full-causal.parquet
 - config_name: full-bidi
-  data_files: [{split: all, path: functional-multiclass-gamba-full-bidi.parquet}]
+  data_files:
+  - split: all
+    path: functional-multiclass-gamba-full-bidi.parquet
 - config_name: 100bp-causal
-  data_files: [{split: all, path: functional-multiclass-gamba-100bp-causal.parquet}]
+  data_files:
+  - split: all
+    path: functional-multiclass-gamba-100bp-causal.parquet
 - config_name: 100bp-bidi
-  data_files: [{split: all, path: functional-multiclass-gamba-100bp-bidi.parquet}]
+  data_files:
+  - split: all
+    path: functional-multiclass-gamba-100bp-bidi.parquet
 ---
 
-# GAMBA functional-region multiclass benchmark
+# GAMBA Functional Region Multiclass
 
-The full configs contain 94,284 rows and pool the retained ROI. The 100 bp
-configs contain 67,626 rows and pool a deterministic BLAKE2-selected 100 bp
-subspan. Filtering `category != "noncoding_regions"` recovers the GAMBA paper
-views with corrected RepeatMasker orientation (84,841 full rows or 58,646
-100 bp rows).
+This representation benchmark asks whether frozen sequence embeddings
+separate genomic functional categories. Each row is one annotated region;
+`label == category`.
 
-Use `*-causal` for left-to-right models and `*-bidi` for
-masked/bidirectional models. The parquet `split` column marks `chr2`, `chr3`,
-`chr16`, and `chr22` as held-out `test`; Hugging Face split `all` loads the
-physical file.
+## Loading
 
-Coordinates are zero-based, half-open hg38. `sequence` is strand-oriented,
-and `pool_start_in_window`/`pool_end_in_window` defines the model evaluation
-span. Repeat rows preserve `repeat_class`; all emitted strands are `+` or `-`.
-The phyloP columns use the pinned Zoonomia 241-mammal track.
+```python
+from datasets import load_dataset
 
-Processing is derived from Microsoft GAMBA commit
-`e83984ea20bb4ee017993144fbe17e7bae3cdddc` and corrects the source
-RepeatMasker strand/class field mapping.
+full_bidi = load_dataset(
+    "Taykhoom/functional-multiclass-gamba",
+    "full-bidi",
+    split="all",
+)
 
-Please cite Consens et al. (2026), doi:10.64898/2026.02.02.703275.
+paper_test = full_bidi.filter(
+    lambda row: row["split"] == "test"
+    and row["category"] != "noncoding_regions"
+)
+```
+
+## Choosing a configuration
+
+The two dimensions are independent:
+
+| Dimension | Options | Meaning |
+|---|---|---|
+| Context | `causal`, `bidi` | End-anchor ROI for autoregressive models or center it for bidirectional models. |
+| Pooling | `full`, `100bp` | Pool over the retained ROI or a deterministic 100 bp subspan. |
+
+The `full` name refers to the retained ROI visible in the 2,048 bp context;
+very long features may be clipped according to GAMBA's context rules. It does
+not mean an unlimited genomic interval.
+
+The 100 bp span is selected deterministically from the ROI. Published
+`pair_id` values remain stable for unchanged source examples, so unrelated
+RepeatMasker corrections do not shift their chosen windows.
+
+## Dataset size
+
+| Config family | Complete rows | Held-out test | Corrected paper rows | Corrected paper test |
+|---|---:|---:|---:|---:|
+| `full-*` | 94,284 | 18,916 | 84,841 | 16,992 |
+| `100bp-*` | 67,626 | 13,593 | 58,646 | 11,781 |
+
+The corrected paper view excludes `category == "noncoding_regions"`.
+
+## Labels
+
+Full configs contain eleven labels:
+
+| Category | Rows |
+|---|---:|
+| repeats | 9,991 |
+| UCNE | 4,169 |
+| vista_enhancer | 2,628 |
+| promoters | 9,971 |
+| UTR5 | 9,802 |
+| UTR3 | 9,693 |
+| coding_regions | 9,893 |
+| exons | 9,915 |
+| introns | 9,026 |
+| upstream_TSS | 9,753 |
+| noncoding_regions | 9,443 |
+
+The 100 bp configs contain 67,626 rows. Promoter entries do not satisfy the
+100 bp retained-span criterion and are absent; `vista_enhancer` has 2,621
+rows, and other categories are filtered by eligible ROI length.
+
+## Evaluation protocol
+
+For the corrected paper evaluation:
+
+1. filter `split == "test"`;
+2. exclude `noncoding_regions`;
+3. pool embeddings over the stored span;
+4. perform cosine leave-one-out 1-nearest-neighbor classification;
+5. report balanced accuracy.
+
+No supervised model-fitting split is implied by Hugging Face split `all`.
+
+## Columns
+
+| Column group | Description |
+|---|---|
+| `split`, `sequence`, `label`, `pair_id` | Chromosome partition, exact input, functional class, and stable feature ID. |
+| `category`, `repeat_class`, `scope`, `context_policy` | Source class, nullable repeat class, `full`/`100bp` scope, and context geometry. |
+| `chrom`, `start`, `end`, `strand` | Zero-based, half-open hg38 feature coordinates. |
+| `context_start`, `context_end` | Forward-genome context coordinates. |
+| `roi_start`, `roi_end` | Retained feature offsets in oriented `sequence`. |
+| `pool_start_in_window`, `pool_end_in_window` | Evaluation span; ROI or selected 100 bp subspan. |
+| `name` | Source annotation identifier. |
+| `phylop_*`, `phylop_context_*` | Pooling-span and symmetric-context phyloP summaries. |
+
+## RepeatMasker correction
+
+RepeatMasker genomic strand and repeat class are stored separately. This
+release corrects an earlier local conversion that placed the repeat class in
+the strand column, then regenerates repeat-derived contexts, pools, controls,
+and phyloP features. Labels are unchanged.
+
+## Processing and citation
+
+Processing and verification:
+
+https://github.com/TaykhoomDalal/Gamba-Processing/tree/main/Functional-Regions
+
+Consens, M. E. et al. *Predicting evolutionary rate as a pretraining task
+improves genome language model representations*. bioRxiv (2026).
+https://doi.org/10.64898/2026.02.02.703275
+
+## License
+
+GAMBA benchmark code/data are MIT licensed. Upstream resources retain their
+own terms.