Convolutional neural network model to predict causal risk factors that share complex regulatory features.

Area of Science:

• Genetics
• Computational Biology
• Bioinformatics

Background:

• Genome-wide association studies (GWASs) have advanced disease genetics.
• Identifying causal noncoding variants with regulatory functions is crucial for post-GWAS analysis.

Purpose of the Study:

• To develop a deep learning framework for modeling complex patterns of risk variants.
• To identify causal noncoding variants with regulatory functions in major psychiatric and autoimmune diseases.

Main Methods:

• Developed a convolutional neural network (CNN) framework utilizing over 2000 functional features.
• Applied the CNN to model combinatorial, nonlinear patterns in risk variants across multiple disease loci.
• Integrated neural and immune features to reflect disease pathophysiology.

Main Results:

• The CNN framework demonstrated high explanatory power for psychiatric and autoimmune diseases.
• Predicted causal variants were located in active regulatory regions and conserved areas.
• Variants were associated with transcription factor interactions and gene expression changes relevant to disease.

Conclusions:

• The developed method effectively identifies potential causal noncoding variants and associated genes.
• This approach aids in the functional interpretation of genetic variants in post-GWAS analyses.
• The findings contribute to a deeper understanding of the genetic basis of complex diseases.