AbstractPrediction in cancer remains limited, and drugs continue to fail clinical trials and post-market validation. The entire multi-ome affects the disease, but typical artificial intelligence and machine learning (AI/ML) cannot generate predictors from multi-omic clinical data. Because, first, the data are “skinny, ” with orders of magnitude more features than samples. Neural networks and deep learning, e.g., would require ≥3B-patient sets to generate models from the 3B-nucleotide genome alone. Second, batch and demographic variations mask the disease-specific patterns. Third, unavoidable imbalances and unavailable labels obscure the distinctions among subpopulations. Fourth, accuracy and interpretability are usually traded off, but (pre)clinical models need to accommodate both. Fifth, the data are of different types, numbers, and dimensions.Our multi-tensor AI/ML is uniquely able to identify accurate, precise, actionable, and mechanistically interpretable predictors from multi-omic clinical data. We have been developing the algorithms, i.e., the “multi-tensor comparative spectral decompositions, ” to extend the mathematics that underlies quantum mechanics to overcome the limitations of typical AI/ML. We demonstrated the algorithms in the discovery and validation of predictors in, e.g., astrocytoma, including glioblastoma (GBM), brain, lung, nerve, ovarian, and uterine cancers. The algorithms identified the predictors repeatedly, in federated and imbalanced public datasets from as few as 50-100 patients each. The GBM predictor, the first to encompass the whole genome and the first to associate a GBM tumor’s DNA copy-number alterations with the patient’s outcome, was additionally prospectively and retrospectively experimentally validated to be the most accurate and precise predictor of survival and response to treatment. Recent experiments also show that the GBM predictor correctly identified previously unrecognized drug targets, i.e., genes that are required for the tumor’s cell proliferation and viability.Here, we demonstrate the multi-tensor AI/ML in the discovery and validation of whole-transcriptome predictors of survival in response to atezolizumab PD-L1 inhibitor immunotherapy. The predictors were discovered in pre-treatment metastatic disease profiles from an open-source clinical trial, and validated in primary disease profiles from the Cancer Genome Atlas (TCGA). One predictor is additionally correlated with the response to the treatment, and the other - with the tissue of metastasis. Kaplan-Meier survival analyses and Cox proportional hazards models show that the two predictors outperform the best indicator of response to atezolizumab to date, i.e., tumor mutation burden (TMB).We conclude that our multi-tensor AI/ML is uniquely suited for personalized medicine.Citation Format:Orly Alter, David B. Oberman, Asaf Zviran. Multi-tensor AI/ML discovery and validation of whole-transcriptome predictors of survival in response to immunotherapy from multi-omic clinical data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3687.
