Navigating the Multiverse: a Hitchhiker's guide to selecting harmonization methods for multimodal biomedical data

Area of Science:

• Biomedical Informatics
• Computational Biology
• Machine Learning

Background:

• Machine learning (ML) significantly enhances biological system comprehension through predictive modeling.
• Multimodal data analysis, integrating diverse data types, shows superior performance over single-modality approaches.
• A lack of comprehensive taxonomy for ML multimodal architectures hinders effective application.

Purpose of the Study:

• To develop a robust framework and taxonomy for multimodal ML analysis.
• To categorize ML architectures for multimodal data, detailing pros and cons.
• To provide a practical guide for selecting and implementing multimodal analysis workflows.

Main Methods:

• Harmonization of multimodal data is presented as a dual process: representation and integration.
• A taxonomy classifies various representation and integration methods into six categories.
• A 10-step guide flowchart is provided for implementing multimodal workflows.

Main Results:

• The study offers a detailed taxonomy of multimodal data harmonization methods.
• Advantages and disadvantages of each method are elucidated.
• A practical, step-by-step guide facilitates the adoption of multimodal approaches.

Conclusions:

• This work provides essential guidance for navigating complex biomedical and clinical data analysis.
• The developed framework supports informed decision-making in multimodal data integration.
• It is a crucial step towards advancing personalized medicine through advanced data analytics.