Expand dataset documentation

README.md

@@ -1,32 +1,135 @@
 ---
 pretty_name: "GAMBA Functional Regions: Feature vs Category-Matched Random"
 license: mit
-tags: [biology, genomics, genome-language-model, benchmark, hg38, parquet]
+tags:
+- biology
+- genomics
+- dna
+- regulatory-genomics
+- genome-language-model
+- benchmark
+- hg38
+- parquet
 configs:
 - config_name: causal
-  data_files: [{split: all, path: functional-random-gamba-causal.parquet}]
+  data_files:
+  - split: all
+    path: functional-random-gamba-causal.parquet
 - config_name: bidi
-  data_files: [{split: all, path: functional-random-gamba-bidi.parquet}]
+  data_files:
+  - split: all
+    path: functional-random-gamba-bidi.parquet
 ---
 
-# GAMBA feature-vs-random benchmark
+# GAMBA Functional Regions: Feature vs Category-Matched Random
 
-Each feature is paired with a chromosome- and length-matched random control
-that avoids retained anchors from the same category. Both configs contain
-188,568 rows; filtering `category != "noncoding_regions"` gives the
-169,682-row corrected paper view.
+This paired binary representation benchmark asks whether a model can
+distinguish an annotated functional region from a chromosome- and
+length-matched random control.
 
-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.
+For this dataset, a random control avoids retained anchors from the **same
+functional category**. It may overlap annotations from other categories.
+Use the annotation-free random dataset if controls must avoid every retained
+annotation category.
 
-Coordinates are zero-based, half-open hg38. `sequence` is strand-oriented,
-`roi_*` marks the feature/control, and `pool_*_in_window` is the evaluation
-span. Repeat rows preserve `repeat_class`; emitted strands are only `+` or
-`-`. The phyloP columns use the pinned Zoonomia 241-mammal track.
+Each feature/control pair shares `pair_id`.
 
-Processing is derived from Microsoft GAMBA commit
-`e83984ea20bb4ee017993144fbe17e7bae3cdddc` and corrects the source
-RepeatMasker strand/class field mapping.
+## Loading
 
-Please cite Consens et al. (2026), doi:10.64898/2026.02.02.703275.
+```python
+from datasets import load_dataset
+
+bidi = load_dataset(
+    "Taykhoom/functional-random-gamba",
+    "bidi",
+    split="all",
+)
+
+paper_test = bidi.filter(
+    lambda row: row["split"] == "test"
+    and row["category"] != "noncoding_regions"
+)
+```
+
+## Dataset size and splits
+
+| Scope | Rows | Held-out `test` rows |
+|---|---:|---:|
+| Complete release | 188,568 | 37,832 |
+| Corrected paper categories | 169,682 | 33,984 |
+
+Labels are balanced: 94,284 `feature` and 94,284 `random` rows.
+
+Rows on `chr2`, `chr3`, `chr16`, and `chr22` are marked `test`; remaining
+included chromosomes are `train`. The Hugging Face split `all` loads the
+whole physical file.
+
+## Categories
+
+The complete release contains 94,284 source features across:
+
+```text
+repeats, UCNE, vista_enhancer, promoters, UTR5, UTR3,
+coding_regions, exons, introns, upstream_TSS, noncoding_regions
+```
+
+`category != "noncoding_regions"` yields the corrected ten-category paper
+view. `repeat_class` is populated on the 19,982 repeat-derived rows
+(feature and matched control).
+
+## Context configurations
+
+| Config | Geometry |
+|---|---|
+| `causal` | ROI end-anchored after strand orientation for autoregressive models. |
+| `bidi` | ROI centered for bidirectional/masked models. |
+
+All sequences are 2,048 bp and use uppercase hg38 bases.
+
+## Evaluation
+
+For the paper protocol:
+
+1. filter `split == "test"`;
+2. exclude `category == "noncoding_regions"`;
+3. pool each representation over
+   `[pool_start_in_window, pool_end_in_window)`;
+4. evaluate cosine leave-one-out 1-nearest-neighbor balanced accuracy.
+
+## Columns
+
+| Column group | Description |
+|---|---|
+| `split`, `sequence`, `label`, `pair_id` | Chromosome partition, exact input, `feature`/`random` class, and matched pair. |
+| `category`, `repeat_class`, `scope` | Source category, nullable repeat class, and `full` scope. |
+| `context_policy` | `causal` or `symmetric`. |
+| `chrom`, `start`, `end`, `strand` | Zero-based, half-open hg38 coordinates. |
+| `context_start`, `context_end` | Forward-genome sequence context coordinates. |
+| `roi_start`, `roi_end` | Feature/control offsets inside oriented `sequence`. |
+| `pool_start_in_window`, `pool_end_in_window` | Evaluation pooling offsets. |
+| `name` | Source/control identifier. |
+| `phylop_*`, `phylop_context_*` | Six ROI and six symmetric-context float32 phyloP summaries. |
+
+## Corrected repeat processing
+
+RepeatMasker genomic strand and repeat class are separate fields. The release
+uses UCSC `rmsk` field 9 as `strand`, field 11 as `repeat_class`, and rebuilds
+repeat-derived controls and sequence orientation. Labels remain unchanged.
+
+Published pair identities and deterministic 100 bp selections for unaffected
+features are preserved.
+
+## 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.