Fix long-range task categorization and seq-length lookup bugs

variant_effect_causal_eqtl was missing from TASK_TO_CATEGORY, so PR #35's
new long-range rows were silently bucketed into "Other". AUPRC (also added
in PR #35) wasn't in CLASSIFICATION_METRICS, so it never surfaced as a
selectable metric. task_seq_lengths.json keys were mixed-case (H2AFZ, ...)
against the lowercased task_name used everywhere else, so histone-mark
task_seq_length was NaN; compute_seq_lengths.py now lowercases on write and
adds the anchor-centered pooling window lengths for the three long-range
tasks, which aren't measurable via the short-range HF dataset it queries.

Also bind the dev server to 0.0.0.0 so it's reachable outside localhost.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

app.py

@@ -555,4 +555,4 @@ with gr.Blocks(title="DNA Benchmark Leaderboard", theme=gr.themes.Soft()) as dem
 
 
 if __name__ == "__main__":
-    demo.launch(share=False)
+    demo.launch(share=False, server_name="0.0.0.0")

src/constants.py

@@ -9,7 +9,9 @@ TASK_CATEGORIES: dict[str, list[str]] = {
     "promoter": ["promoter_all", "promoter_tata", "promoter_no_tata"],
     "enhancer": ["enhancers", "enhancers_types"],
     "splice_site": ["splice_sites_all", "splice_sites_donors", "splice_sites_acceptors"],
-    "variant_effect": ["variant_effect_pathogenic_clinvar", "variant_effect_pathogenic_omim"],
+    "variant_effect": [
+        "variant_effect_pathogenic_clinvar", "variant_effect_pathogenic_omim", "variant_effect_causal_eqtl",
+    ],
     "rna_expression": ["bulk_rna_expression"],
 }
 
@@ -29,6 +31,7 @@ CLASSIFICATION_METRICS: dict[str, str] = {
     "accuracy_test": "Accuracy",
     "weighted_f1_test": "Weighted F1",
     "macro_f1_test": "Macro F1",
+    "auprc_test": "AUPRC",
 }
 
 CATEGORY_DISPLAY_NAMES: dict[str, str] = {

tools/compute_seq_lengths.py

@@ -16,7 +16,7 @@ import os
 import pathlib
 from typing import Any
 
-from datasets import DatasetDict, load_dataset
+from datasets import load_dataset
 from dotenv import load_dotenv
 
 load_dotenv()
@@ -25,36 +25,37 @@ HF_TOKEN = os.environ.get("HF_TOKEN")
 HF_DATASET = "InstaDeepAI/nucleotide_transformer_downstream_tasks_revised"
 SAMPLE_SIZE = 100
 
+# Names match the HF dataset's own data_dir casing (e.g. "H2AFZ"), not the lowercased
+# task_name used elsewhere in the dash — see the .lower() when writing results below.
 TASKS = [
     "H2AFZ", "H3K27ac", "H3K27me3", "H3K36me3",
     "H3K4me1", "H3K4me2", "H3K4me3", "H3K9ac", "H3K9me3", "H4K20me1",
     "promoter_all", "promoter_tata", "promoter_no_tata",
     "enhancers", "enhancers_types",
     "splice_sites_all", "splice_sites_donors", "splice_sites_acceptors",
-    "variant_effect_pathogenic_clinvar", "variant_effect_pathogenic_omim",
-    "bulk_rna_expression",
 ]
 
-OUTPUT_PATH = pathlib.Path(__file__).parent / "task_seq_lengths.json"
-
-
-def load_dataset_for_task(hf_dataset: str, task_name: str) -> DatasetDict:
-    """Load HuggingFace dataset for specific task.
-    Directly from the bdna package
-    """
-    kwargs: dict[str, Any] = {}
-    if HF_TOKEN:
-        kwargs["token"] = HF_TOKEN
+# variant_effect_causal_eqtl, variant_effect_pathogenic_clinvar, and bulk_rna_expression are
+# long-range tasks (100,000bp anchor-centered windows, see PR #35 in bench-dna). They don't
+# exist in HF_DATASET so they can't be measured the same way as the short-range tasks above,
+# and their raw window length (100,000bp for all of them) wouldn't be a meaningful x-axis value
+# anyway. These are the anchor-centered pooling windows the Genomics LRB paper actually reads
+# from that window -- the portion of the sequence that's relevant to the label.
+LONG_RANGE_SEQ_LENGTHS: dict[str, int] = {
+    "variant_effect_causal_eqtl": 1536,
+    "variant_effect_pathogenic_clinvar": 1536,
+    "bulk_rna_expression": 383 + 256,
+}
 
-    if hf_dataset == "InstaDeepAI/genomics-long-range-benchmark":
-        return load_dataset(hf_dataset, task_name=task_name, **kwargs)
-
-    return load_dataset(hf_dataset, data_dir=task_name, **kwargs)
+OUTPUT_PATH = pathlib.Path(__file__).parent / "task_seq_lengths.json"
 
 
 def compute_mean_length(task_name: str) -> int | None:
     try:
-        ds_dict = load_dataset_for_task(HF_DATASET, task_name)
+        kwargs: dict[str, Any] = {}
+        if HF_TOKEN:
+            kwargs["token"] = HF_TOKEN
+        ds_dict = load_dataset(HF_DATASET, data_dir=task_name, **kwargs)
         ds = ds_dict["train"]
         # Sample a subset to keep it fast
         sample = ds.select(range(min(SAMPLE_SIZE, len(ds))))
@@ -80,7 +81,9 @@ def main():
     for task in TASKS:
         length = compute_mean_length(task)
         if length is not None:
-            result[task] = length
+            result[task.lower()] = length
+
+    result.update(LONG_RANGE_SEQ_LENGTHS)
 
     OUTPUT_PATH.write_text(json.dumps(result, indent=2))
     print(f"\nWrote {len(result)} entries to {OUTPUT_PATH}")

tools/task_seq_lengths.json

@@ -1,14 +1,14 @@
 {
-  "H2AFZ": 1000,
-  "H3K27ac": 1000,
-  "H3K27me3": 1000,
-  "H3K36me3": 1000,
-  "H3K4me1": 1000,
-  "H3K4me2": 1000,
-  "H3K4me3": 1000,
-  "H3K9ac": 1000,
-  "H3K9me3": 1000,
-  "H4K20me1": 1000,
+  "h2afz": 1000,
+  "h3k27ac": 1000,
+  "h3k27me3": 1000,
+  "h3k36me3": 1000,
+  "h3k4me1": 1000,
+  "h3k4me2": 1000,
+  "h3k4me3": 1000,
+  "h3k9ac": 1000,
+  "h3k9me3": 1000,
+  "h4k20me1": 1000,
   "promoter_all": 300,
   "promoter_tata": 300,
   "promoter_no_tata": 300,
@@ -16,5 +16,8 @@
   "enhancers_types": 400,
   "splice_sites_all": 600,
   "splice_sites_donors": 600,
-  "splice_sites_acceptors": 600
+  "splice_sites_acceptors": 600,
+  "variant_effect_causal_eqtl": 1536,
+  "variant_effect_pathogenic_clinvar": 1536,
+  "bulk_rna_expression": 639
 }