Drug Repurposing Candidates for Triple-Negative Breast Cancer

Builds on 'Mechanism of Welch+BH anti-conservativeness at 4-vs-N': re-run the 9-vs-9 within-breast screen with exact permutation p-values (>=2000 perms/drug, chunked streaming, float32) plus BH-FDR, produce the corrected q-value ranking, and quantify how many of the 7 iteration-2 hits survive; additionally fit subtype-corrected models (AUC ~ TNBC + basal/ER/PR/HER2 from the subtype table) for each surviving hit to quantify the basal-lineage confound, and preregister the mk-1775 claim with fixed line assignments and permutation statistics

Builds on 'Jackknife sign-flip law for 4-line drug selectivity statistics' (pattern hunter): compute exact leave-one-out sign-flip fractions for n_TNBC = 5–26 by subsampling breast and pan-cancer lines across all GDSCv2 drugs, fit logit(P_flip) = a(n) + b(n)|t| per n, state one precise closed-form conjecture for b(n) and the minimum stable-n, validate on held-out n values and drugs not used for fitting, and derive an analytic Gaussian benchmark P_flip(n,|t|) to test whether the empirical law deviates from the textbook prediction (deviation = heavy-tail/heteroscedasticity signature)

Builds on the abandoned DepMap triage and 'PRISM counter-screen structurally failed' (skeptic + closer of open items): attach the researcher-downloaded DepMap Chronos matrix via chunked loading, run the pan-essential gene filter, and test whether WEE1 or any hit target is a TNBC-selective dependency after filtering; dock the surviving hits' ligands to their targets with vina as triage; then attack the lead's and explorer's claims — attempt counterexamples against the corrected permutation ranking (label-permutation stress, alternative n choices) and against the n-scaling law (search for violating subsample configurations), reporting any claim that breaks