A self-supervised learning framework for discovering cortical folding patterns under genetic influence: Application...

Area of Science:

• Neuroscience
• Genetics
• Computational Biology

Background:

• Cortical folding patterns offer insights into human brain development.
• High inter-individual variability complicates the association of these patterns with developmental pathologies.
• The anterior cingulate cortex, particularly the paracingulate sulcus, is of psychiatric interest.

Purpose of the Study:

• To propose a framework for discovering candidate cortical folding patterns influenced by genetics.
• To apply this framework to the anterior cingulate cortex region.
• To identify novel genetic loci associated with specific cortical folding patterns.

Main Methods:

• Utilized a self-supervised deep learning algorithm on 36,000 UK Biobank subjects for regional fold variability representation.
• Trained linear classification models to discern and generalize folding patterns.
• Applied the framework to identify genetic loci associated with anterior cingulate cortex folding patterns.

Main Results:

• Generalization of paracingulate sulcus labeling in the UK Biobank cohort did not yield clear genetic associations.
• The framework successfully discovered 4 loci in the right hemisphere and 10 loci in the left hemisphere associated with anterior cingulate cortex folding patterns (p < 5x10^-8).
• One locus was replicated in a smaller, non-white British ancestry cohort, and many discovered loci show prior associations with brain anatomy or psychiatry.

Conclusions:

• The developed framework is effective in discovering novel genetic loci associated with specific cortical folding patterns.
• While direct genetic associations for paracingulate sulcus shape were not found, the method highlights potential genetic underpinnings of anterior cingulate cortex folding.
• The findings contribute to understanding the genetic architecture of brain development and its relation to psychiatric conditions.