Abstract Inferring gene module activity is key to understanding transcriptomic dysregulation in case-control studies. Most existing gene set analysis methods focus on differences between groups without considering the complex geometry of the data. We introduce GEDO, a graph and topology-based method that infers gene module activity through a transition score, quantifying the shift from healthy controls to diseased individuals. When applied to bulk RNA-seq data from Sjögren’s disease patients and healthy controls (PRECISESADS cohort), and a breast cancer dataset from The Cancer Genome Atlas, GEDO was benchmarked against Principal Component Analysis (PCA), the mean of z-scores, Single Sample Gene Set Enrichment Analysis (ssGSEA), and Gene Set Variation Analysis (GSVA) in classification, unsupervised clustering tasks and robustness against noise and bias. On the PRECISESADS cohort, GEDO outperformed the other approaches in predicting disease status, interferon signature, enhancing subgroup separability, and robustness against noise and bias. The biological signal captured was aligned with the knowledge and clinical features of Sjögren’s disease. In the breast cancer dataset, GEDO’s embeddings better represent PAM50 molecular subtypes. Its supervised and topology-based design enables finer resolution of disease-related transcriptomic alterations. GEDO offers a robust, interpretable framework for quantifying gene modules’ activity, with applications in single and potentially in multi-omics integration.
Bézier et al. (Thu,) studied this question.