GAUDI: interpretable multi-omics integration with UMAP embeddings and density-based clustering
Pol Castellano‐Escuder, Derek K. Zachman, Kevin Han, Matthew D. Hirschey
Abstract
Integrating high-dimensional cellular multi-omics data is crucial for understanding various layers of biological control. Single ‘omic methods provide important insights, but often fall short in handling the complex relationships between genes, proteins, metabolites and beyond. Here, we present a novel, non-linear, and unsupervised method called GAUDI (Group Aggregation via UMAP Data Integration) that leverages independent UMAP embeddings for the concurrent analysis of multiple data types. GAUDI uncovers non-linear relationships among different omics data better than several state-of-the-art methods. This approach not only clusters samples by their multi-omic profiles but also identifies latent factors across each omics dataset, thereby enabling interpretation of the underlying features contributing to each cluster. Consequently, GAUDI facilitates more intuitive, interpretable visualizations to identify novel insights and potential biomarkers from a wide range of experimental designs. Integration of multi-omics data is challenging due to high dimensionality and non-linear relationships. Here, authors develop an unsupervised method that leverages UMAP embeddings and density-based clustering to integrate diverse omics data types and identifies biologically meaningful patterns across multiple benchmarks.