High-dimensional multi-study multi-modality covariate-augmented generalized factor model.

Area of Science:

• Statistics
• Machine Learning
• Bioinformatics

Background:

• Latent factor models are crucial for integrating data from multiple sources.
• Existing methods struggle to integrate both multi-study and multi-modality data simultaneously.
• There is a need for flexible models that handle diverse data types across studies.

Purpose of the Study:

• To develop a high-dimensional generalized factor model for integrating multi-modality data from multiple studies.
• To investigate identifiability conditions for enhanced model interpretability.
• To address computational challenges in high-dimensional nonlinear integration.

Main Methods:

• Introduced a high-dimensional generalized factor model accommodating covariates.
• Employed a variational lower bound approximation for the observed log-likelihood.
• Utilized M-estimation theory and a variational expectation-maximization (EM) algorithm for parameter estimation.
• Developed a criterion for determining the optimal number of study-shared and study-specific factors.

Main Results:

• The proposed model effectively integrates multi-modality data across multiple studies.
• Identifiability conditions were established, improving model interpretability.
• The variational EM algorithm demonstrated computational efficiency.
• The method significantly outperformed existing approaches in simulation studies and a real-world application regarding accuracy and speed.

Conclusions:

• The novel generalized factor model offers a powerful solution for integrated analysis of multi-study, multi-modality data.
• The method provides accurate parameter estimation and computational efficiency.
• This approach advances latent factor modeling for complex, heterogeneous datasets.