Simultaneous Representation Learning of Multi-Omics and Clinical Outcome Data via a Supervised Knowledge-Guided...

Area of Science:

• Computational Biology
• Genomics
• Biostatistics

Background:

• High-throughput techniques generate vast multi-omics and clinical data for diseases.
• Simultaneous representation learning of high-dimensional, heterogeneous multi-modality data with clinical outcomes is challenging.
• Existing methods have limitations in integrating diverse data types for biological insight.

Purpose of the Study:

• To develop a supervised knowledge-guided Bayesian factor model for integrative analysis of multi-omics and clinical outcome data.
• To simultaneously extract informative low-dimensional representations and predict clinical outcomes.
• To identify active biological modalities and features for meaningful structural identification of high-dimensional data.

Main Methods:

• A novel supervised knowledge-guided Bayesian factor model is proposed.
• The model utilizes two-level adaptive shrinkage in hierarchical priors for feature and modality selection.
• It incorporates robustness to noisy biological graph data and handles continuous and categorical data types.

Main Results:

• The proposed method successfully extracts low-dimensional representations and predicts clinical outcomes.
• It effectively identifies active modalities and features, leading to biologically meaningful data structures.
• Analyses of Alzheimer's disease (AD) data, including multi-omics and imaging, reveal advantages over existing methods.

Conclusions:

• The developed Bayesian model provides a powerful tool for integrative multi-omics and clinical data analysis.
• It offers meaningful insights into the biological mechanisms of complex diseases, exemplified by Alzheimer's disease.
• The method's robustness and ability to handle diverse data types enhance its applicability in biomedical research.