Integrating multimodal data through interpretable heterogeneous ensembles.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Machine Learning

Background:

• Multimodal data integration is crucial for predicting biomedical outcomes.
• Existing methods often fail to capture unique information within individual data modalities.
• Late integration approaches, while promising, are understudied in biomedicine.

Approach:

• Propose Ensemble Integration (EI), a systematic late integration framework.
• EI builds local predictive models from individual modalities and integrates them using ensemble algorithms.
• A novel interpretation method for EI models is also introduced.

Key Points:

• EI significantly enhances prediction accuracy compared to single modalities.
• EI outperforms established early integration methods in biomedical applications.
• EI successfully predicted protein function and COVID-19 mortality using multimodal data.

Conclusions:

• Ensemble Integration provides an effective framework for biomedical data integration.
• EI's ability to leverage diverse data sources improves predictive modeling.
• The approach offers interpretable models with identified disease-relevant features.