What is Latent Omics Learning?

Latent Omics Learning (LOL) is a framework for evaluating phenotype prediction from high-dimensional omics data while explicitly examining information leakage and metadata-associated confounding. LOL combines leakage-safe preprocessing, imbalance-aware evaluation, group-aware validation, the Leakage Inflation Index (ΔLII), and PVCA-lite diagnostics. LOL provides a reproducible analytical layer that places prediction, validation, and diagnostic assessment within the same workflow.

Latent Omics Learning (LOL) is available as a Python package and analytical framework for leakage-aware phenotype prediction and confounding diagnostics in transcriptomics datasets. It allows researchers to evaluate whether apparently strong phenotype-prediction results are robust or may be influenced by information leakage, repeated samples, tissue differences, class imbalance, or metadata-associated confounding.

Preprint: Muigano MN. LOL: a Python package for leakage-aware phenotype prediction and confounding diagnostics in GEO transcriptomics datasets. bioRxiv, 2026. DOI: 10.64898/2026.09.26.754596.