Interpretable machine learning for low-sample multi-omics: a case study of ferret vaccine response

Area of Science:

• Computational Biology
• Systems Biology
• Immunology

Background:

• Machine learning models are crucial for biological pathway analysis but often lack interpretability.
• Interpretable machine learning (IML) aims to bridge the gap between predictive accuracy and biological understanding.

Purpose of the Study:

• To apply an IML framework to multi-omics data for identifying molecular signatures of vaccine response.
• To demonstrate that IML can provide biologically meaningful insights without compromising predictive performance.

Main Methods:

• Utilized an IML framework combining TreeFARMS and Rashomon Set analysis on ferret multi-omics data (transcriptomics, proteomics, lipidomics, metabolomics).
• Generated sparse, interpretable decision trees optimized for accuracy and compactness.
• Identified stable and alternative molecular rules predictive of vaccination status.

Main Results:

• The IML models outperformed ensemble methods in predicting vaccination status.
• Concise if-then rules linked specific molecular features (e.g., AZGP1 expression, ketoleucine levels) to immune response patterns.
• Successfully captured biologically meaningful signatures of vaccine response.

Conclusions:

• IML provides a transparent and reproducible alternative to black-box models in multi-omics analysis.
• This proof-of-concept demonstrates the utility of IML for uncovering mechanisms of vaccine response.
• The developed framework can enhance understanding of complex biological systems and accelerate vaccine development.