| import numpy as np | |
| import pandas as pd | |
| from scipy import stats | |
| from common import RES, hom, axes, centre, effects, sgd_labels | |
| THRESHOLDS = [200, 250, 300, 350, 380, 400, 418] | |
| base = None | |
| rows = [] | |
| for t in THRESHOLDS: | |
| Z = hom(min_nonmissing=t) | |
| ag, ox, os_, ref = axes(Z.columns) | |
| Zc = centre(Z) | |
| e_ox, e_os = effects(Zc, ox, os_) | |
| cd = (e_ox - e_os).abs().dropna() | |
| oi, si = np.where(ox)[0], np.where(os_)[0] | |
| a = Zc.iloc[:, oi[0::2]].mean(axis=1) - Zc.iloc[:, si[0::2]].mean(axis=1) | |
| b = Zc.iloc[:, oi[1::2]].mean(axis=1) - Zc.iloc[:, si[1::2]].mean(axis=1) | |
| m = a.notna() & b.notna() | |
| if t == 350: | |
| base = cd | |
| lab = sgd_labels("oxidative") | |
| lab = lab[lab.index.isin(e_ox.dropna().index)] | |
| conc = (np.sign(e_ox.loc[lab.index]) == -lab).mean() | |
| rows.append({"min_nonmissing": t, "genes": len(cd), "splithalf_cd_signed": stats.spearmanr(a[m], b[m]).statistic, | |
| "splithalf_cd_abs": stats.spearmanr(a[m].abs(), b[m].abs()).statistic, | |
| "oxidative_benchmark_n": len(lab), "oxidative_concordance": conc, "cd": cd}) | |
| for r in rows: | |
| top_r = set(r["cd"].sort_values(ascending=False).index[: int(np.ceil(0.1 * len(r["cd"])))]) | |
| top_b = set(base.sort_values(ascending=False).index[: int(np.ceil(0.1 * len(base)))]) | |
| r["top_decile_jaccard_vs_350"] = len(top_r & top_b) / len(top_r | top_b) | |
| del r["cd"] | |
| out = pd.DataFrame(rows) | |
| out.to_csv(RES / "threshold_sensitivity.csv", index=False) | |