Helpers for comparing biomarker preprocessing pipelines on the same cohort under leakage-free cross-validation. Includes:
os_log_transform_adaptive()— per-feature pseudocount log2 transform that survives mixtures of raw counts and pre-normalized ratios.os_plate_median_frozen— R6 class for fold-frozen plate-median (or batch-median) centering, simpler and more deployable than ComBat when only a location shift is suspected.os_paired_delong()— paired DeLong test wrapper for AUC comparison of two scoring functions on the same cohort.os_oof_pipeline_compare()— grouped-CV driver that runs each pipeline train-fold-frozen, pools out-of-fold predictions per seed, and reports paired DeLong tests vs a reference pipeline.
Details
These functions exist to make the "is method X better than the current
pipeline?" question answerable in a single call instead of an ad-hoc script.
They are meant to compose with existing OmicSelector primitives such as
clr_transform, FrozenComBat, ws_rank,
os_panel_null_benchmark, and os_clustered_bootstrap_auc.
References
DeLong ER, DeLong DM, Clarke-Pearson DL. (1988). Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics, 44(3), 837-845.
Robin X, Turck N, Hainard A, et al. (2011). pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics, 12, 77.
Stouffer SA, Suchman EA, DeVinney LC, Star SA, Williams RM. (1949). The American Soldier: Adjustment During Army Life. Princeton University Press.