Learning heritable multimodal brain representation via contrastive learning.

Area of Science:

• Neuroimaging
• Genetics
• Machine Learning

Background:

• Magnetic resonance imaging (MRI)-derived phenotypes (IDPs) have facilitated the discovery of genomic loci linked to brain structure and function.
• Existing IDPs often rely on single imaging modalities, potentially limiting comprehensive genetic discovery by overlooking complementary cross-modal information.

Purpose of the Study:

• To introduce a multimodal contrastive learning framework for deriving heritable brain representations from paired T1- and T2-weighted MRIs.
• To improve the scope of genetic discovery by integrating information across multiple imaging modalities.

Main Methods:

• Developed a momentum-based contrastive learning framework utilizing paired T1- and T2-weighted MRI data.
• Derived heritable representations from multimodal brain imaging data.

Main Results:

• The learned representations demonstrated enhanced correlation with traditional IDPs and improved prediction of age and brain disorders.
• Genome-wide association studies (GWAS) of the learned representations revealed significantly higher overlap of genetic loci across modalities.
• Identified shared protein and drug targets from GWAS loci, providing biological insights.

Conclusions:

• The multimodal contrastive learning framework effectively learns shared brain representations across imaging modalities.
• These representations exhibit improved anatomical and genetic coherence, advancing multimodal neuroimaging and genetic discovery.