Remics: a redescription-based framework for multi-omics analysis

Complex diseases such as cancer are characterized by their intricate etiology, arising from several molecular mechanisms that span multiple omic layers. To obtain insights on disease subtypes, associated biomarkers, and improve prognostic modeling, it is essential to integrate and interpret multi-omics data in a biologically meaningful way. We introduce Remics, a redescription-based framework for multi-omics integration inspired by higher-order statistical representations. Remics leverages higher-order cumulants to identify redescriptions, which are sets of multi-omics features that jointly capture equivalent biological variation across modalities. These feature groups are further analyzed through network representations, multi-omics risk scoring, and biomarker discovery to reveal molecular interactions underlying disease mechanisms. We applied Remics on simulated data as well as multi-omics data of six different cancer types from The Cancer Genome Atlas. We demonstrate that redescription-based integration uncovers functionally coherent cross-omics feature associations and compare them with state-of-the-art approaches. Our results highlight the potential of higher-order multi-omics statistical analysis to advance precision medicine through improved interpretability and discovery of novel molecular relationships.