NegBioDB / tests /test_ge_llm_dataset.py
jang1563's picture
NegBioDB final: 4 domains, fully audited
6d1bbc7
"""Tests for GE LLM dataset builder."""
import json
from pathlib import Path
import pandas as pd
import pytest
from negbiodb_depmap.llm_dataset import (
apply_max_per_gene,
assign_splits,
construct_evidence_description,
construct_l3_context,
construct_l4_context,
write_jsonl,
write_dataset_metadata,
)
@pytest.fixture
def sample_row():
return pd.Series({
"gene_symbol": "TP53",
"model_id": "ACH-000001",
"ccle_name": "A549_LUNG",
"lineage": "Lung",
"primary_disease": "Non-Small Cell Lung Cancer",
"gene_effect_score": 0.05,
"dependency_probability": 0.10,
"evidence_type": "crispr_nonessential",
"source_db": "depmap",
"confidence_tier": "gold",
"is_common_essential": 0,
"is_reference_nonessential": 1,
"gene_degree": 1500,
})
class TestConstructEvidence:
def test_basic_fields(self, sample_row):
text = construct_evidence_description(sample_row)
assert "TP53" in text
assert "A549_LUNG" in text
assert "Lung" in text
assert "CRISPR" in text
def test_gene_effect_included(self, sample_row):
text = construct_evidence_description(sample_row)
assert "0.050" in text
def test_dependency_prob_included(self, sample_row):
text = construct_evidence_description(sample_row)
assert "0.100" in text
def test_rnai_evidence(self, sample_row):
sample_row["evidence_type"] = "rnai_nonessential"
text = construct_evidence_description(sample_row)
assert "RNAi" in text
class TestConstructL3:
def test_includes_gene_function(self, sample_row):
text = construct_l3_context(sample_row, gene_description="Tumor suppressor")
assert "Tumor suppressor" in text
def test_includes_reference_flag(self, sample_row):
text = construct_l3_context(sample_row)
assert "non-essential gene set" in text
def test_includes_degree(self, sample_row):
text = construct_l3_context(sample_row)
assert "1500" in text
class TestConstructL4:
def test_minimal_context(self, sample_row):
text = construct_l4_context(sample_row)
assert "TP53" in text
assert "A549_LUNG" in text
assert "Lung" in text
# Should NOT include scores
assert "0.050" not in text
class TestApplyMaxPerGene:
def test_caps_per_gene(self):
df = pd.DataFrame({
"gene_id": [1] * 10 + [2] * 5,
"value": range(15),
})
result = apply_max_per_gene(df, max_per_gene=3)
assert len(result[result["gene_id"] == 1]) == 3
assert len(result[result["gene_id"] == 2]) == 3
def test_no_change_below_max(self):
df = pd.DataFrame({
"gene_id": [1, 2, 3],
"value": [10, 20, 30],
})
result = apply_max_per_gene(df, max_per_gene=5)
assert len(result) == 3
class TestAssignSplits:
def test_all_assigned(self):
df = pd.DataFrame({"x": range(100)})
result = assign_splits(df)
assert "split" in result.columns
assert all(result["split"].isin(["train", "val", "test"]))
def test_approximate_ratios(self):
df = pd.DataFrame({"x": range(1000)})
result = assign_splits(df)
counts = result["split"].value_counts()
assert 650 < counts["train"] < 750
assert 50 < counts["val"] < 150
class TestWriteJsonl:
def test_writes_file(self, tmp_path):
records = [{"id": 1, "text": "hello"}, {"id": 2, "text": "world"}]
path = tmp_path / "test.jsonl"
count = write_jsonl(records, path)
assert count == 2
assert path.exists()
# Verify content
lines = path.read_text().strip().split("\n")
assert len(lines) == 2
assert json.loads(lines[0])["text"] == "hello"
class TestWriteMetadata:
def test_writes_metadata(self, tmp_path):
write_dataset_metadata(
tmp_path, task="ge-l1",
stats={"n_records": 100, "split_counts": {"train": 70, "val": 10, "test": 20}},
)
meta_path = tmp_path / "ge-l1_metadata.json"
assert meta_path.exists()
meta = json.loads(meta_path.read_text())
assert meta["task"] == "ge-l1"
assert meta["domain"] == "ge"
assert meta["n_records"] == 100