CADD-Splice-improving genome-wide variant effect prediction using deep learning-derived splice scores.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Accurate human protein synthesis relies on mRNA splicing.
• Genetic variants impacting splicing cause many diseases but are hard to detect.
• Deep neural networks (DNNs) show promise in predicting variant effects on splicing.

Purpose of the Study:

• To compare machine learning methods for predicting variant effects on splicing.
• To integrate effective splicing prediction methods into genome-wide variant effect predictors.
• To evaluate the impact on classifying known pathogenic variants.

Main Methods:

• Compared several machine learning methods using experimental data for splicing variant effects.
• Integrated top-performing splicing prediction approaches into genome-wide models.
• Utilized CADD (Combined Annotation Dependent Depletion) framework, developing CADD-Splice.

Main Results:

• Integration of DNN-based splicing scores into CADD significantly improved predictions.
• CADD-Splice enhanced classification of variants across multiple categories without performance loss.
• Specialized splice effect scores outperform general predictors for splice variants.

Conclusions:

• Integrating specialized molecular process scores, like splicing DNNs, improves genome-wide variant effect prediction.
• The CADD-Splice model demonstrates a successful approach for enhancing variant interpretation.
• This strategy is generalizable to other molecular processes for improved genetic disease prediction.