Annotating functional effects of non-coding variants in neuropsychiatric cell types by deep transfer learning

Area of Science:

• Genomics and Computational Biology
• Neuroscience and Genetics

Background:

• Genome-wide association studies (GWAS) identify numerous loci linked to neuropsychiatric traits, but understanding their molecular mechanisms is challenging.
• Existing computational tools for variant annotation struggle with under-represented neurodevelopmental cell types, hindering functional interpretation.

Purpose of the Study:

• To develop an advanced computational tool, MetaChrom, for accurate functional annotation of DNA variants in neurodevelopmental contexts.
• To prioritize genetic variants associated with neuropsychiatric disorders using deep learning and epigenomic data.

Main Methods:

• Trained deep Convolutional Neural Networks (ResNet) using a comprehensive dataset of neurodevelopment-related cell and tissue types.
• Employed transfer learning by integrating public epigenomic data to enhance model accuracy.
• Compared MetaChrom's performance against established variant annotation tools like CADD and delta-SVM for chromatin accessibility prediction.

Main Results:

• MetaChrom demonstrated significantly superior performance in predicting experimentally validated chromatin accessibility variants compared to existing tools.
• Integration of GWAS data with MetaChrom predictions successfully prioritized 31 single nucleotide polymorphisms (SNPs) associated with Schizophrenia.
• Identified potential risk genes and their relevant biological contexts for Schizophrenia.

Conclusions:

• MetaChrom provides crucial functional annotations for DNA variants within the neurodevelopmental context.
• The MetaChrom methodology offers a generalizable approach that can be extended to annotate variants in other disease-specific cell or tissue types.
• This tool aids in deciphering the functional impact of genetic variations on neuropsychiatric disorders.