Machine Learning Prediction of Non-Coding Variant Impact in Cell-Class-Specific Human Retinal Cis-Regulatory Elements

Area of Science:

• Genomics
• Computational Biology
• Ophthalmology

Background:

• Non-coding variants in cis-regulatory elements (CREs) are implicated in inherited retinal diseases (IRDs).
• Functional characterization of these regulatory variants is a significant challenge in genetic research.
• Identifying disease-associated variants requires robust predictive tools.

Purpose of the Study:

• To develop machine learning (ML) models for predicting the functional impact of non-coding variants on retinal CREs.
• To enhance the identification and prioritization of variants relevant to IRDs.
• To leverage single nucleus ATAC-seq data for cell-class-specific variant impact prediction.

Main Methods:

• Implemented a gapped k-mer support vector machine (SVM) approach.
• Trained 18 distinct ML models using single nucleus ATAC-seq data from human retina cell classes.
• Predicted variant impact on 39,437 cell-class-specific regulatory elements.

Main Results:

• ML models achieved over 90% accuracy with high cell class specificity.
• Variant Impact Prediction (VIP) scores identified specific sequences within CREs, including transcription factor (TF) binding motifs, sensitive to mutation.
• VIP scores demonstrated predictive value in massively parallel reporter assays for single nucleotide variants and indels.

Conclusions:

• Single nucleus epigenomic data can effectively predict the functional impact of non-coding sequence variants.
• The developed ML models and VIP scores enable rapid prioritization of patient variants for functional analysis.
• This approach advances the understanding of genetic contributions to IRDs.