TPClust: Temporal Profile-Guided Subtyping Using High-Dimensional Omics Data

Clustering is widely used to identify subtypes in heterogeneous populations, yet most approaches rarely integrate longitudinal phenotypic trajectories with high-dimensional molecular profiles, limiting their ability to resolve biologically and clinically meaningful heterogeneity in progressive diseases. We developed TPClust, a supervised, semi-parametric clustering method that integrates high-dimensional omics data with longitudinal phenotypes including outcomes and covariates for outcome-guided subtyping. TPClust jointly models latent subtype membership and longitudinal outcome trajectories using multinomial logistic regression informed by molecular features selected via structured regularization, along with spline-based regression to capture subtype-specific, time-varying covariate effects. Simulations demonstrate valid inference for time-varying effects and robust feature selection. Applied to transcriptomic profiles and longitudinal cognitive data from 1,020 older adults in the Religious Orders Study and the Rush Memory and Aging Project, TPClust identified four aging subtypes including intermediate subtypes not captured by unimodal approaches with distinct cognitive trajectories, time-varying risk profiles, clinical and neuropathological features, and multimodal molecular signatures.