Ensemble INtegration Suite – EINS Manuscript Datasets (Ensemble Multi-Omics Integration Increases Robustness and Interpretability)
Year: 2026
Type: article
Abstract: Integrating multi-omics data such as genomics, proteomics, metabolomics, and lipidomics provides deeper insights into cellular biology and disease than single-omics analyses. The diversity of available integration strategies, each based on distinct statistical frameworks, often produces heterogeneous, difficult-to-compare results. We developed Ensemble INtegration Suite (EINS), a framework that combines outputs from multiple multi-omics integration methods to support sample subtyping and biomarker exploration. We evaluated EINS using data from Cancer Cell Line Encyclopedia, with 313 cell lines characterized by DNA methylation, miRNA expression, transcriptomics, proteomics, and metabolomics. Sample subtyping was assessed with five unsupervised evaluations. Our ensemble Median Partition Analysis (Ens-MPA) performed best overall. Supervised analyses tested representation separability, i.e. how well the integrated representations separated labeling, for which we considered cell line primary sites and, within a single primary-site, its subtypes. We used six classifiers: Random Forest, SVM, XGBoost, Logistic Regression, LASSO, and Elastic Net. We determined improvements in macro-averaged AUC over single omics and multi-omics integration. We applied EINS to proteomics and lipidomics of lipid transfer protein knockout cell lines. Ensemble feature analysis identified protein-lipid groupings associated with shared biological functions and offered new insights. We believe that ensemble integration provides a more reliable approach for uncovering multi-omics molecular patterns. EINS is available at https://github.com/EnsembleMultiomics/EINS.
Authors: Zuiderwijk, Mara, Mohammed, Yassene
Institution: Leiden University Medical Center
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7225308167