Fig. 1: GAUDI method overview and workflow.
From: GAUDI: interpretable multi-omics integration with UMAP embeddings and density-based clustering
The GAUDI (Group Aggregation via UMAP Data Integration) method integrates multi-omics data through a non-linear unsupervised approach. The workflow consists of five steps: (1) Independent UMAP embedding is applied to each omics dataset, transforming matrices of different dimensionalities (n×p1, n×p2, etc.) into standardized representations of equal dimensions (n×d); (2) These embeddings are concatenated to create a combined data matrix; (3) A second UMAP transformation is applied to the concatenated matrix to create an integrated representation; (4) Hierarchical Density-Based Spatial Clustering (HDBSCAN) is used to identify sample groups with similar multi-omic profiles; and (5) XGBoost combined with SHAP values is used to calculate feature importance, enabling biological interpretation of the factors driving sample clustering. This approach mitigates bias from high-dimensional omics types, captures non-linear relationships between variables, and provides interpretable results that can reveal novel biological insights across diverse experimental designs.
