Self-Normalizing Deep Learning for Enhanced Multi-Omics Data Analysis in Oncology

Investigating multi-omics data is crucial for unraveling the complex biological mechanisms underlying cancer, thereby enabling effective strategies for prevention, early detection, diagnosis, and treatment. However, predicting patient outcomes through the integration of all available multi-omics data remains an understudied topic. We present SeNMo, a self-normalizing deep neural network trained on multi-omics data across 33 cancer types. SeNMo is particularly efficient in handling multi-omics data characterized by high-width (many features) and low-length (fewer samples) attributes. We trained SeNMo for predicting overall survival of patients using pan-cancer multi-omics data involving 33 cancer sites from the publicly available NCI Genomics Data Commons. The multi-omics data includes gene expression, DNA methylation, miRNA expression, DNA mutations, protein expression modalities, and clinical data. SeNMo was validated on two independent cohorts: Moffitt Cancer Center and CPTAC lung squamous cell carcinoma. The baseline model, trained to predict overall survival, achieved a concordance index (C-Index) of 0.76 in the validation regime and 0.758 on the held-out test set. Furthermore, the model demonstrated strong generalizability, achieving an average accuracy of 99.8% in predicting primary cancer type across pan-cancer cohort. SeNMo further demonstrated significant performance (p< 0.05) in predicting tertiary lymph structures from multi-omics data, showing generalizability across cancer types, molecular data types, and clinical endpoints. SeNMo and similar models are poised to transform the oncology landscape, offering hope for more effective, efficient, and personalized cancer care.