Clustering longitudinal multivariate trajectories using an ensemble of principal component trees.

Area of Science:

• Biomedical data analysis
• Computational biology
• Statistical modeling

Background:

• Clustering longitudinal data presents challenges due to high dimensionality, irregular sampling, and noise.
• Identifying meaningful temporal subgroups and their defining features is crucial for biomedical research.

Purpose of the Study:

• To introduce TAPIO, a flexible and interpretable framework for longitudinal data clustering.
• To enable identification of temporal subgroups and their key features in complex datasets.

Main Methods:

• TAPIO is an ensemble-based clustering approach with longitudinal extensions (longTAPIO_trajectories, longTAPIO_sample, longTAPIO_MLD).
• It integrates dimension reduction and cluster-specific feature importance for robust clustering.
• Accommodates univariate/multivariate, regularly/irregularly sampled longitudinal data.

Main Results:

• Simulations show TAPIO accurately recovers cluster structure and identifies relevant features.
• longTAPIO_trajectories performs well on regularly sampled data; longTAPIO_MLD excels with irregular measurements.
• Applications reveal patient subgroups with distinct survival patterns and molecularly distinct proteomics clusters.

Conclusions:

• TAPIO provides a scalable, interpretable framework for longitudinal clustering of complex, high-dimensional data.
• It aids in identifying meaningful clusters and their defining features.
• Potential applications include patient stratification, biomarker discovery, and advancing longitudinal data analysis.