Integrating Neuroimaging and Genetics via Contrastive Learning for Working Memory.

Area of Science:

• Cognitive Neuroscience
• Genetics
• Neuroimaging

Background:

• Understanding working memory is vital for cognitive neuroscience and identifying cognitive impairment biomarkers, especially in aging populations.
• Existing research often analyzes genetic and neural data separately, limiting comprehensive understanding.

Purpose of the Study:

• To integrate single nucleotide polymorphism (SNP) and neuroimaging data for improved working memory capacity classification.
• To identify genetic factors associated with brain structure and function in working memory.

Main Methods:

• Utilized UK Biobank data, integrating 1060 SNPs from Alzheimer's disease-related protein networks with latent brain gray matter density features extracted by a Convolutional Neural Network (CNN).
• Employed supervised contrastive learning to enhance genetic-imaging relationships within individuals and working memory groups.
• Applied sparse canonical correlation analysis to uncover significant genetic variants and brain region associations.

Main Results:

• The contrastive learning model outperformed baseline methods in classifying high vs. low working memory capacity.
• Identified significant associations between genetic variants in genes like FYN, RPL28, and MAPT, enriched in dendrite and synapse pathways, and brain regions including the cerebellum and striatum.

Conclusions:

• The study provides novel insights into the genetic and neural underpinnings of working memory.
• Findings highlight the role of specific genes and brain structures in cognitive function and may inform future research and therapeutic strategies for cognitive impairments.