Ultra-fast variant effect prediction using biophysical transcription factor binding models.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Sequence variations in transcription factor (TF)-binding sites impact gene expression and disease.
• Current deep learning models struggle with accurate variant effect prediction, particularly for common genetic variants.
• Biophysical models of TF binding are crucial for interpreting variant effects and gaining mechanistic insights.

Purpose of the Study:

• Introduce motifDiff, a novel computational tool for quantifying variant effects on TF-DNA interactions.
• Enhance the accuracy and interpretability of variant effect prediction using biophysical models.
• Provide a scalable and statistically rigorous method for analyzing millions of genetic variants.

Main Methods:

• Developed motifDiff using mono- and dinucleotide position weight matrices.
• Implemented a statistically rigorous normalization strategy for optimal performance.
• Evaluated motifDiff on diverse ground truth datasets of in vivo common variant effects.

Main Results:

• motifDiff demonstrates high efficacy in predicting variant effects across multiple datasets.
• The tool provides scalable analysis, scoring millions of variants rapidly.
• Established robust benchmarks for variant effect prediction accuracy.

Conclusions:

• motifDiff offers a significant advancement in predicting the functional impact of genetic variants.
• The tool enhances mechanistic understanding of TF-DNA interactions and gene regulation.
• motifDiff provides unique insights into human accelerated regions and variant interpretation.