Analyzing associations and higher-order effects in multi-omics data with double machine learning

Integrative omics analyses enhance our understanding of disease mechanisms and biomarkers by investigating relationships among traits, omics measurements, genetic variants, and epidemiological factors. Statistically, these analyses are challenging and require robust, flexible methodologies due to high dimensionality, non-standard data distributions, and potentially complex, non-linear confounding effects.

To facilitate the integration and analysis of multi-omics data, we introduce the Robust Omics MethodologY (ROMY) framework and its corresponding R implementation, the romy package. ROMY enables users to (a) perform robust association testing between two target variables while incorporating flexible covariate adjustments, (b) examine effects on measurement variances and covariances (e.g., co-expression, co-abundance), and (c) conduct rigorous interaction-effect testing. ROMY builds on recent advances in theoretical statistics and double machine learning to ensure robustness and statistical validity. We illustrate the performance of our framework through simulation studies.