Joint multi-omics discriminant analysis with consistent representation learning using PANDA.

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• Integrative multi-omics analysis offers deeper biological insights than single omics.
• Existing methods struggle with inconsistent data distributions and balancing correlation vs. discrimination.

Purpose of the Study:

• To introduce PAN-omics Discriminant Analysis (PANDA), a novel joint discriminant analysis method.
• To address limitations in current multi-omics integration techniques.

Main Methods:

• PANDA jointly learns consistent discriminant latent representations for each omics dataset.
• It maximizes between-class and minimizes within-class variations in a common space.
• The method models relationships at consistency representation and cross-omics correlation levels.

Main Results:

• PANDA effectively minimizes distribution differences among omics, yielding robust latent representations.
• It overcomes the compromise between correlation and discrimination seen in other methods.
• PANDA outperformed 10 state-of-the-art methods on simulated and real-world datasets.

Conclusions:

• PANDA provides a robust and effective framework for integrative multi-omics analysis.
• The method enhances disease modeling by improving data integration and representation learning.
• Available in R and MATLAB, PANDA offers a valuable tool for researchers.