Multi-study factor analysis.

Area of Science:

• Multivariate statistics
• Bioinformatics
• Genomics

Background:

• Standard factor analysis methods are typically applied to single datasets.
• Combining results from multiple studies often requires ad-hoc approaches, limiting comprehensive analysis.
• Identifying shared and unique patterns across studies is crucial for robust biological insights.

Purpose of the Study:

• To introduce a novel factor analysis methodology for the joint analysis of multiple studies.
• To develop methods for distinguishing common factors (shared across studies) from study-specific factors.
• To provide a computational tool for accelerating cross-study unsupervised analysis and assessing signal reproducibility.

Main Methods:

• Developed a novel class of factor analysis methodologies for joint multi-study analysis.
• Utilized an Expectation Conditional-Maximization (ECM) algorithm for parameter estimation.
• Created a procedure for selecting the optimal number of common and specific factors.
• Applied the method to gene expression data in ovarian cancer.

Main Results:

• The proposed method successfully identifies and estimates both common and study-specific factors.
• Simulations demonstrate the method's effectiveness in evaluating performance.
• Application to ovarian cancer gene expression data highlights benefits over standard factor analysis.
• The developed R package (MSFA) facilitates joint unsupervised analysis across studies.

Conclusions:

• The novel factor analysis approach enables robust joint analysis of multiple studies.
• This methodology improves the understanding of cross-study signal reproducibility in multivariate datasets.
• The MSFA R package provides a valuable tool for researchers in bioinformatics and genomics.