Exploring machine learning strategies for predicting cardiovascular disease risk factors from multi-omic data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Machine learning (ML) classifiers are increasingly used for predicting cardiovascular disease (CVD) and related risk factors using omics data.
• Challenges include categorical outcomes and class imbalances, with limited understanding of factors influencing prediction quality.

Purpose of the Study:

• To compare different machine learning strategies for predicting CVD risk factors.
• To evaluate the influence of ML classifiers, omics data types, and dimension reduction techniques on prediction quality.

Main Methods:

• Compared six ML classifiers using blood-derived metabolomics, epigenetics, and transcriptomics data.
• Utilized unsupervised and semi-supervised autoencoders for omics dimension reduction.
• Constructed multi-omic predictions using a meta-learner and evaluated performance using F1 score.
• Investigated transfer learning with pre-trained autoencoders on an external cohort.

Main Results:

• Multi-omics predictions generally outperformed single-omics predictions, especially for extreme risk factor levels.
• Semi-supervised autoencoders enhanced downstream predictions compared to unsupervised methods.
• Transfer learning demonstrated median Area Under the Curve gains of 0.09-0.14 for transcriptomics and 0.07-0.11 for metabolomics.

Conclusions:

• The study provides a framework for researchers to assess the impact of omics data, ML classifiers, and dimension reduction on CVD risk factor prediction.
• Highlights the benefit of multi-omics integration and advanced dimension reduction techniques for improved predictive performance.