Genetic Variant Effect Prediction of Major Depressive Disorder Using Large Language Models

Mental health issues are critical global public health concerns. Current methods for exploring genetic variants linked to mental illnesses often can’t pinpoint causal variants and are resource intensive. This project explores the Genomic Pre-trained Network Multiple Sequence Alignment (GPN-MSA) model for predicting genetic variant effects associated with Major Depressive Disorder (MDD). An MSA of a human genome with 100 vertebrate species and a dataset of Single Nucleotide Polymorphisms (SNPs) linked to MDD were input into the GPN-MSA. The model calculated a variant effect prediction (VEP) score for each variant. SNPs significant for MDD had more negative VEP scores than less significant ones, suggesting GPN-MSA may be able to identify deleterious variants associated with MDD. The GPN-MSA also assigned extreme negative scores to previously overlooked SNPs, indicating their potential association with MDD. This study successfully tested the GPN-MSA for predicting MDD variant effects.