Machine Learning Prediction of Non-Coding Variant Impact in Human Retinal cis-Regulatory Elements

Area of Science:

• Genomics
• Computational Biology
• Ophthalmology

Background:

• Cis-regulatory variants are implicated in retinal diseases, but their functional impact is difficult to determine.
• Accurate prediction of regulatory variant function is crucial for understanding genetic contributions to vision disorders.

Purpose of the Study:

• To develop and validate a machine learning model for predicting the functional impact of cis-regulatory variants in the human retina.
• To systematically quantify the predicted impact of single nucleotide variants within retinal cis-regulatory elements.

Main Methods:

• Utilized human retinal DNA accessibility data (ATAC-seq) to identify cis-regulatory elements.
• Trained a gapped k-mer support vector machine model on 80% of identified elements.
• Applied in silico saturation mutagenesis and variant scoring, validated against population frequency, conservation, TF binding, and reporter assay data.

Main Results:

• Achieved 95% accuracy in distinguishing retinal regulatory elements from negative sequences.
• Scored all single nucleotide variants within a hold-out set of retinal CREs.
• Negative impact scores correlated with increased conservation, disrupted TF motifs, and reduced reporter expression.

Conclusions:

• Demonstrated the effectiveness of machine learning trained on retinal epigenomic data for predicting non-coding variant impact.
• The model accurately scores sequences and predicts transcription factor binding motifs.
• This approach can accelerate the identification of pathogenic variants in unexplained retinal diseases.