Bioinformatics Advances

Across 9 methods and 2 omics datasets, batch effects degraded nearly all of them, with random forest and MICE most stable and mean imputation most fragile.

Benchmarked against six published tools across four negative-sampling and data-split settings, it outperformed them in all four, including on unseen epitopes, where most others struggled.

Concatenation raised revealable rank over the summed matrix in all four datasets, with spliced and unspliced factors enriching for distinct nascent and steady-state programs.

maasMDS calibrates ancestry-specific distances across genotyping arrays before embedding, so cohorts merge without collapsing to shared SNPs. In Indigenous American cohorts, it traced minority European segments to their colonial sources and resolved geographic clines across Mexico and Central America that imputation and Procrustes alignment obscured.

A strict contract retained just 44 of 173 resolved SugarBind glycans; WURCS led Recall@5 (0.688), but a popularity control scored higher on MRR and average precision.

This paper introduces scfdr, an R package that applies an empirical Bayes framework to threshold sparse correlation matrices by controlling the false discovery rate among retained non-zero correlations, rather than minimizing the Frobenius norm. Across simulated data and real glioma and glioblastoma gene-expression datasets, this FDR-based approach retained more true correlations than the Frobenius-minimizing RSC method, with greater sensitivity for moderate correlations.