Resources

Traditional patient stratification relies on individual biomarkers or supervised outcome models. State-based stratification groups patients by their position in a landscape of biological programs — differentiation stage, immune context, regulatory networks — derived from unsupervised multi-omics analysis. This approach captures biological heterogeneity that single markers miss and enables more robust, generalizable stratification across cohorts.

Outcome-supervised models risk encoding the biases of their training data — treatment protocols, follow-up durations, population demographics. Outcome-agnostic models capture intrinsic disease biology first, then allow outcomes to be mapped onto the biological landscape as an independent validation layer. This separation ensures biological fidelity, prevents information leakage, and produces representations that generalize across studies and institutions.