An integrated Bayesian framework for multi-omics prediction and classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Multi-omics data integration is crucial for discovering clinically actionable biomarkers.
• Existing methods for phenotype prediction from omics data often lack integration of longitudinal, multi-modal information.
• Decentralized frameworks for analyzing time-dependent omics data are limited.

Purpose of the Study:

• To propose a novel Bayesian ensemble method for consolidating predictions from multiple longitudinal and cross-sectional omics data layers.
• To enable uncertainty quantification and interval estimation in predictions derived from integrated omics data.
• To address the limitations of existing frequentist approaches in handling complex, time-dependent multi-omics datasets.

Main Methods:

• Developed a Bayesian ensemble approach to integrate information across diverse omics data types (longitudinal and cross-sectional).
• Implemented a method that allows for uncertainty quantification in predictions.
• Utilized posterior summaries for interval estimation of key quantities.

Main Results:

• Successfully applied the method to four published multi-omics datasets.
• Demonstrated recapitulation of known biological insights and discovery of novel findings.
• Showcased superior performance over existing methods in estimation, prediction, and uncertainty quantification.

Conclusions:

• The proposed Bayesian ensemble method effectively integrates multi-omics data for improved biomarker discovery and disease prediction.
• The approach provides robust uncertainty quantification, outperforming current methodologies.
• Open-source software is available, facilitating broader application in biomedical research.