"""Tests for PPI LLM dataset utilities (src/negbiodb_ppi/llm_dataset.py).""" import json from pathlib import Path import numpy as np import pandas as pd import pytest from negbiodb_ppi.llm_dataset import ( DETECTION_METHOD_DESCRIPTIONS, JSONL_SCHEMA_FIELDS, MAX_PER_PROTEIN, SOURCE_TO_L1_CATEGORY, apply_max_per_protein, assign_splits, construct_evidence_description, construct_l3_context, construct_l4_context, read_jsonl, write_jsonl, ) # ── SOURCE_TO_L1_CATEGORY Tests ────────────────────────────────────────── class TestSourceToL1Category: def test_intact_maps_to_a(self): assert SOURCE_TO_L1_CATEGORY["intact_gold"] == "A" assert SOURCE_TO_L1_CATEGORY["intact_silver"] == "A" def test_huri_maps_to_b(self): assert SOURCE_TO_L1_CATEGORY["huri"] == "B" def test_humap_maps_to_c(self): assert SOURCE_TO_L1_CATEGORY["humap"] == "C" def test_string_maps_to_d(self): assert SOURCE_TO_L1_CATEGORY["string"] == "D" def test_four_categories(self): assert set(SOURCE_TO_L1_CATEGORY.values()) == {"A", "B", "C", "D"} # ── construct_evidence_description Tests ───────────────────────────────── class TestConstructEvidenceDescription: def test_intact_easy(self): rec = {"source_db": "intact", "detection_method": "co-immunoprecipitation", "gene_symbol_1": "TP53", "gene_symbol_2": "CDK2"} desc = construct_evidence_description(rec, "easy") assert "co-immunoprecipitation" in desc assert "TP53" in desc assert "CDK2" in desc assert "No physical interaction" in desc def test_huri_easy(self): rec = {"source_db": "huri", "gene_symbol_1": "TP53", "gene_symbol_2": "CDK2"} desc = construct_evidence_description(rec, "easy") assert "yeast two-hybrid" in desc.lower() or "Y2H" in desc def test_humap_easy(self): rec = {"source_db": "humap", "gene_symbol_1": "X", "gene_symbol_2": "Y"} desc = construct_evidence_description(rec, "easy") assert "machine learning" in desc.lower() or "co-fractionation" in desc.lower() def test_string_easy(self): rec = {"source_db": "string", "gene_symbol_1": "X", "gene_symbol_2": "Y"} desc = construct_evidence_description(rec, "easy") assert "evidence channels" in desc.lower() or "score" in desc.lower() def test_difficulty_changes_wording(self): rec = {"source_db": "intact", "detection_method": "co-immunoprecipitation", "gene_symbol_1": "X", "gene_symbol_2": "Y"} easy = construct_evidence_description(rec, "easy") hard = construct_evidence_description(rec, "hard") assert easy != hard # Different wording for different difficulties # ── construct_l3_context Tests ─────────────────────────────────────────── class TestConstructL3Context: def test_includes_both_proteins(self): rec = { "gene_symbol_1": "TP53", "uniprot_1": "P04637", "seq_len_1": 393, "function_1": "Tumor suppressor", "location_1": "Nucleus", "domains_1": "p53 domain", "gene_symbol_2": "INS", "uniprot_2": "P01308", "seq_len_2": 110, "function_2": "Insulin", "location_2": "Extracellular", "domains_2": "Insulin domain", "detection_method": "co-immunoprecipitation", } ctx = construct_l3_context(rec) assert "TP53" in ctx assert "INS" in ctx assert "P04637" in ctx assert "393" in ctx assert "Tumor suppressor" in ctx assert "Insulin" in ctx assert "co-immunoprecipitation" in ctx def test_handles_missing_fields(self): rec = {"gene_symbol_1": "X", "gene_symbol_2": "Y"} ctx = construct_l3_context(rec) assert "X" in ctx assert "Y" in ctx # ── construct_l4_context Tests ─────────────────────────────────────────── class TestConstructL4Context: def test_minimal_context(self): rec = {"gene_symbol_1": "TP53", "gene_symbol_2": "BRCA1"} ctx = construct_l4_context(rec) assert "TP53" in ctx assert "BRCA1" in ctx assert "Homo sapiens" in ctx assert "tested" in ctx.lower() # ── apply_max_per_protein Tests ────────────────────────────────────────── class TestApplyMaxPerProtein: def test_caps_at_max(self): df = pd.DataFrame({ "protein_id_1": [1] * 20 + [2] * 5, "protein_id_2": [3] * 25, "value": range(25), }) result = apply_max_per_protein(df, max_per_protein=10) # Protein 1 appears 20 times in column 1 → capped counts_p1 = (result["protein_id_1"] == 1).sum() assert counts_p1 <= 10 def test_no_op_under_limit(self): df = pd.DataFrame({ "protein_id_1": [1, 1, 2, 2, 3], "protein_id_2": [4, 5, 6, 7, 8], "value": range(5), }) result = apply_max_per_protein(df, max_per_protein=10) assert len(result) == 5 def test_reproducibility(self): df = pd.DataFrame({ "protein_id_1": [1] * 50, "protein_id_2": [2] * 50, "value": range(50), }) r1 = apply_max_per_protein(df, max_per_protein=10, rng=np.random.RandomState(42)) r2 = apply_max_per_protein(df, max_per_protein=10, rng=np.random.RandomState(42)) pd.testing.assert_frame_equal(r1, r2) def test_default_max(self): assert MAX_PER_PROTEIN == 10 # ── assign_splits Tests ────────────────────────────────────────────────── class TestAssignSplits: def test_correct_split_sizes(self): df = pd.DataFrame({"x": range(100)}) result = assign_splits(df, fewshot_size=10, val_size=20, test_size=70, seed=42) assert (result["split"] == "fewshot").sum() == 10 assert (result["split"] == "val").sum() == 20 assert (result["split"] == "test").sum() == 70 assert len(result) == 100 def test_reproducibility(self): df = pd.DataFrame({"x": range(50)}) r1 = assign_splits(df, 5, 10, 35, seed=42) r2 = assign_splits(df, 5, 10, 35, seed=42) pd.testing.assert_frame_equal(r1, r2) def test_undersized_dataset(self): df = pd.DataFrame({"x": range(20)}) result = assign_splits(df, fewshot_size=10, val_size=5, test_size=100, seed=42) assert len(result) == 20 assert (result["split"] == "fewshot").sum() == 10 assert (result["split"] == "val").sum() == 5 assert (result["split"] == "test").sum() == 5 # ── JSONL I/O Tests ────────────────────────────────────────────────────── class TestJSONLIO: def test_write_read_roundtrip(self, tmp_path): records = [ {"question_id": "PPIL1-001", "task": "ppi-l1", "gold_answer": "A"}, {"question_id": "PPIL1-002", "task": "ppi-l1", "gold_answer": "B"}, ] path = tmp_path / "test.jsonl" write_jsonl(records, path) loaded = read_jsonl(path) assert loaded == records def test_empty_write(self, tmp_path): path = tmp_path / "empty.jsonl" write_jsonl([], path) loaded = read_jsonl(path) assert loaded == [] def test_unicode_handling(self, tmp_path): records = [{"name": "α-synuclein", "function": "Sjögren's protein"}] path = tmp_path / "unicode.jsonl" write_jsonl(records, path) loaded = read_jsonl(path) assert loaded[0]["name"] == "α-synuclein" assert loaded[0]["function"] == "Sjögren's protein" # ── Schema Field Tests ─────────────────────────────────────────────────── class TestSchemaFields: def test_required_fields_present(self): for field in ["question_id", "task", "split", "gold_answer", "context_text"]: assert field in JSONL_SCHEMA_FIELDS def test_gold_answer_not_correct_answer(self): """PPI uses gold_answer (consistent with CT domain).""" assert "gold_answer" in JSONL_SCHEMA_FIELDS assert "correct_answer" not in JSONL_SCHEMA_FIELDS # ── Detection Method Descriptions Tests ────────────────────────────────── class TestDetectionMethodDescriptions: def test_has_common_methods(self): assert "co-immunoprecipitation" in DETECTION_METHOD_DESCRIPTIONS assert "pull down" in DETECTION_METHOD_DESCRIPTIONS assert "two hybrid" in DETECTION_METHOD_DESCRIPTIONS def test_descriptions_are_readable(self): for method, desc in DETECTION_METHOD_DESCRIPTIONS.items(): assert len(desc) > 5, f"Description for {method} too short"