KRAS mutations drive approximately 25% of human cancers, resulting in 3.9 million new cases annually. Recently approved KRAS inhibition (KRASi) monotherapies demonstrate limited efficacy and durability. Diverse combination therapies are under investigation. This necessitates biomarker strategies for identifying patients most likely to respond to KRASi monotherapy or combination therapy.The Genialis™ Supermodel is a foundation model of cancer biology that can be used to derive novel biomarkers for diagnosis, clinical development, and treatment decision-making in oncology. Trained on nearly 1 million harmonized transcriptomic samples, this large molecular model (LMM) maps individual patient RNA-seq samples into biologically meaningful spaces defined by distinct combinations of biological signatures. Each biological subspace reflects mechanisms relevant to a specific therapeutic class. These mappings enable the selection of core biological features, which are then used in downstream machine-learning models to predict therapeutic response and stratify patients.The Supermodel captures biology linked to responses to KRASi, EGFR inhibitors, ICI (immune checkpoint inhibitors), and standard-of-care chemotherapy, among other therapeutic mechanisms, generating classifiers that predict monotherapy responses in diverse preclinical and clinical cohorts. We further demonstrate that the intersection of these individual classifiers identifies clinical contexts where monotherapy is sufficient and where combination therapies may provide additional benefit. These results refine patient stratification and guide clinical decision-making by aligning treatment selection with the underlying tumor biology and therapeutic context.The Genialis™ Supermodel establishes a novel and robust framework for precision oncology, enhancing patient stratification, optimizing clinical decision-making, and expanding opportunities for effective KRAS-driven cancer treatments.