Probabilistic pathway-based multimodal factor analysis.

Area of Science:

• Biomedical data analysis
• Computational biology
• Systems biology

Background:

• Multimodal profiling integrates diverse biological data for deeper insights.
• Current analytical strategies struggle with low sample numbers and interpretability.
• Factor analysis in molecular biology often lacks direct biological interpretation.

Purpose of the Study:

• To develop a novel multimodal factor analysis approach for pathway-level interpretation.
• To create a method that integrates information from various profiling technologies.
• To enable the derivation of concrete biological hypotheses from complex datasets.

Main Methods:

• Developed PathFA, a Bayesian multimodal factor analysis approach operating on pathways.
• PathFA is efficient, hyper-parameter free, and infers observation noise automatically.
• Combines pathway-learning with integrative multimodal analysis.

Main Results:

• PathFA demonstrates strong performance on small sample sizes and real tumor data (proteomics and transcriptomics).
• Successfully recovered pathway activity associated with poor prognosis in melanoma patients.
• Identified pathways linked to specific cell types and tumor heterogeneity.

Conclusions:

• PathFA provides integrative and interpretable views across multimodal profiling data.
• The method captures known biology, making it suitable for analyzing multimodal sample cohorts.
• PathFA facilitates hypothesis generation and understanding of complex biological systems.