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

Area of Science:

• Biostatistics
• Genomics
• Neuroscience

Background:

• Clustering methods often fail to integrate longitudinal data with high-dimensional omics, limiting subtype discovery in progressive diseases.
• Resolving heterogeneity in aging requires methods that combine molecular profiles with evolving phenotypic trajectories.

Purpose of the Study:

• To develop and validate TPClust, a novel supervised, semi-parametric clustering approach for outcome-guided subtyping.
• To integrate high-dimensional omics data with longitudinal phenotypes for enhanced disease heterogeneity analysis.

Main Methods:

• TPClust employs multinomial logistic regression for latent subtype membership and longitudinal outcome trajectories.
• Structured regularization selects molecular features, while spline regression models time-varying covariate effects.
• Simulations confirmed valid inference for time-varying effects and robust feature selection.

Main Results:

• TPClust identified four distinct aging subtypes in 1,020 older adults using transcriptomic and cognitive data.
• These subtypes exhibited unique cognitive trajectories, time-varying risk profiles, and neuropathological features.
• Novel intermediate subtypes, missed by unimodal methods, were discovered.

Conclusions:

• TPClust effectively integrates multimodal data to reveal complex aging heterogeneity.
• The identified subtypes offer deeper insights into disease progression and personalized risk stratification.
• This method advances the understanding of biologically and clinically meaningful subtypes in aging populations.