MOINER: A Novel Multiomics Early Integration Framework for Biomedical Classification and Biomarker Discovery

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multiomics data integration is crucial for precision medicine, disease diagnosis, and biomarker discovery.
• Early integration via matrix concatenation is simple but suffers from information loss and limited feature interaction.

Purpose of the Study:

• To present a novel multiomics early integration framework (MOINER).
• To address limitations of traditional early integration methods by enhancing information and utilizing image representation learning.

Main Methods:

• Developed MOINER, a framework incorporating information enhancement and image representation learning.
• Employed a self-attention mechanism to capture intrinsic correlations within omics features.

Main Results:

• MOINER significantly outperforms existing state-of-the-art methods in multiomics data integration.
• The framework provides interpretable insights through attention embedding visualization and potential biomarker identification.

Conclusions:

• MOINER offers an advanced approach to multiomics data integration.
• The framework enhances precision medicine by improving biological process understanding and facilitating biomarker discovery.