#!/usr/bin/env python3 """Unit and public-source integration tests for the additive sensitivity audit.""" import json import random import unittest import run_sensitivity as analysis class UnitTests(unittest.TestCase): def test_frozen_threshold_grid(self): protocol = json.loads((analysis.HERE / "protocol.json").read_text()) self.assertEqual(analysis.THRESHOLDS, protocol["gse65145"]["reciprocal_overlap_thresholds"]) def test_censoring_is_not_exact_numeric(self): for cell, expected in [("<0.2", "left_censored"), (">0.2", "right_censored"), ("≤0.2", "left_censored"), ("≥0.2", "right_censored"), (None, "missing"), ("NA", "missing"), ("inf", "nonfinite")]: value, status = analysis.numeric_cell(cell) self.assertIsNone(value) self.assertEqual(status, expected) self.assertEqual(analysis.numeric_cell("0.2"), (0.2, "observed")) def test_median_paired_delta_not_difference_of_medians(self): rows = [{"gentamicin_percent": g, "untreated_percent": u, "gentamicin_status": "observed", "untreated_status": "observed", "context_6nt": str(i)} for i, (g, u) in enumerate([(0, 0), (10, 9), (11, 0)])] result = analysis.ptc_estimates(rows) self.assertEqual(result["median_paired_delta_percentage_points"], 1) self.assertEqual(result["difference_of_marginal_medians_percentage_points"], 10) self.assertEqual(result, analysis.ptc_estimates(list(reversed(rows)))) rows[0]["gentamicin_status"] = "left_censored" self.assertIn("insufficient", analysis.ptc_estimates(rows)["status"]) self.assertIn("insufficient", analysis.ptc_estimates([])["status"]) def test_coordinate_selection_no_signal_input(self): anchor = {"chrom": "chr1", "txStart": 100, "txEnd": 200} coordinates = [("chr1", 99, 201), ("chr1", 100, 200), ("chr1", 0, 1000), ("chr2", 100, 200)] self.assertEqual(analysis.eligible_feature_indices(coordinates, anchor, .95), [1, 0]) self.assertEqual(analysis.eligible_feature_indices(coordinates, anchor, 1), [1]) self.assertEqual(analysis.eligible_feature_indices([coordinates[0]], anchor, 1), []) def test_geometric_positive_guard_and_no_pseudocount(self): values = {"Pro": [1, 4], "Qui": [1, 1], "Rap": [2, 8]} self.assertAlmostEqual(analysis.contrasts(values, "geometric")["ratios"]["Rap_over_Pro"], 2) values["Qui"][0] = 0 self.assertIn("insufficient", analysis.contrasts(values, "geometric")["status"]) values["Qui"] = [] self.assertIn("insufficient", analysis.contrasts(values)["status"]) class SourceIntegrationTests(unittest.TestCase): @classmethod def setUpClass(cls): cls.sources = analysis.checked_sources(analysis.ROOT) cls.ptc = analysis.ptc_analysis(analysis.ROOT) cls.gse = analysis.gse_analysis(analysis.ROOT) def test_verified_sources_and_original_scripts(self): self.assertEqual(len(self.sources), 14) def test_ptc_inventory_and_censoring(self): self.assertEqual(self.ptc["complete_source_table_inventory"]["populated_source_rows"], 5837) self.assertEqual(self.ptc["complete_source_table_inventory"]["blank_formatted_rows"], 5837) self.assertEqual(self.ptc["valid_UGA_rows_in_complete_table"], 2724) self.assertEqual(self.ptc["nearest_variant_row_count"], 12) self.assertEqual(self.ptc["unique_nearest_6nt_context_count"], 8) self.assertEqual(self.ptc["minimum_hamming_distance"], 1) self.assertEqual(self.ptc["exact_6nt_rows"], 0) for counts in self.ptc["value_status_counts"].values(): self.assertEqual(counts["observed"], 12) self.assertEqual(sum(v for k, v in counts.items() if k != "observed"), 0) def test_ptc_finite_leave_outs_and_group_removal(self): rows = self.ptc["nearest_rows"] baseline = self.ptc["baseline"] shuffled = rows.copy() random.Random(24).shuffle(shuffled) # Test order-invariance only; analysis has no stochastic step. self.assertEqual(baseline, analysis.ptc_estimates(shuffled)) self.assertAlmostEqual(baseline["median_paired_delta_percentage_points"], .0165) self.assertEqual(baseline["positive_paired_deltas"], 11) self.assertEqual(baseline["negative_paired_deltas"], 1) self.assertEqual(len(self.ptc["leave_one_variant_row_out"]), 12) self.assertTrue(all(r["n_variant_rows"] == 11 for r in self.ptc["leave_one_variant_row_out"])) self.assertEqual(len(self.ptc["leave_one_context_group_out"]), 8) for row in self.ptc["leave_one_context_group_out"]: context = row["omitted_context_6nt"] expected_omitted = {r["excel_row"] for r in rows if r["context_6nt"] == context} self.assertEqual(set(row["omitted_excel_rows"]), expected_omitted) self.assertEqual(row["n_unique_contexts"], 7) self.assertEqual(row["n_variant_rows"], 12 - len(expected_omitted)) ranges = [("row_leave_out_median_paired_delta_range_percentage_points", .014, .019), ("context_group_leave_out_median_paired_delta_range_percentage_points", .0135, .0285)] for key, low, high in ranges: self.assertAlmostEqual(self.ptc[key]["min"], low) self.assertAlmostEqual(self.ptc[key]["max"], high) def test_gse_rows_thresholds_and_fixed_n(self): self.assertEqual(self.gse["sample_counts_per_condition"], {"Pro": 2, "Qui": 2, "Rap": 2}) self.assertEqual(self.gse["row_count_per_track"], 149122) self.assertEqual(self.gse["locus_overlapping_row_count"], 18) self.assertEqual([r["qualifying_feature_count"] for r in self.gse["threshold_summaries"]], [3, 3, 3, 3, 0]) self.assertEqual(self.gse["threshold_summaries"][-1]["status"], "insufficient_no_eligible_feature") for f in self.gse["features"]: self.assertEqual(len(f["leave_one_sample_out"]), 6) for row in f["leave_one_sample_out"]: self.assertEqual(sorted(row["remaining_sample_counts"].values()), [1, 2, 2]) for ratios in f["cross_sample_ratios"].values(): self.assertEqual(len(ratios["values"]), 4) def test_gse_primary_and_direction_robustness(self): primary = self.gse["primary_feature"] self.assertEqual(primary["track_row_1based_without_header"], 116255) self.assertAlmostEqual(primary["arithmetic"]["ratios"]["Rap_over_Pro"], .5857063642506453) self.assertAlmostEqual(primary["geometric"]["ratios"]["Rap_over_Pro"], .584998, places=6) for f in self.gse["features"]: self.assertLess(f["arithmetic"]["ratios"]["Rap_over_Pro"], 1) self.assertLess(f["geometric"]["ratios"]["Rap_over_Pro"], 1) self.assertTrue(all(r["ratios"]["Rap_over_Pro"] < 1 for r in f["leave_one_sample_out"])) self.assertLess(f["cross_sample_ratios"]["Rap_over_Pro"]["max"], 1) self.assertLess(f["arithmetic"]["ratios"]["Qui_over_Pro"], 1) self.assertGreater(f["arithmetic"]["ratios"]["Rap_over_Qui"], 1) def test_bounded_machine_readable_interpretation(self): result = analysis.interpretations(self.ptc, self.gse) self.assertEqual(result["stable"][0]["status"], "positive_in_all_deletions") self.assertEqual(result["stable"][1]["status"], "below_one_in_all_calculated_diagnostics") self.assertEqual(result["insufficient"][0]["status"], "insufficient_no_eligible_feature") self.assertEqual(result["insufficient"][1]["status"], "not_identifiable_from_these_data") if __name__ == "__main__": unittest.main(verbosity=2)