MetaRNN: differentiating rare pathogenic and rare benign missense SNVs and InDels using deep learning

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Computational methods are crucial for understanding genetic variants.
• Current methods struggle to differentiate rare pathogenic from rare benign variants.

Purpose of the Study:

• To develop advanced pathogenicity prediction models for rare genetic variants.
• To improve the identification and prioritization of nonsynonymous single nucleotide variants (nsSNVs) and non-frameshift insertion/deletions (nfINDELs).

Main Methods:

• Utilized context annotations and deep learning techniques.
• Developed two models: MetaRNN for nsSNVs and MetaRNN-indel for nfINDELs.
• Employed independent test sets for model validation.

Main Results:

• MetaRNN and MetaRNN-indel demonstrated superior performance compared to state-of-the-art methods.
• Achieved a more interpretable score distribution for pathogenicity predictions.
• Prediction scores from both models were comparable, facilitating integrated analyses.

Conclusions:

• MetaRNN and MetaRNN-indel offer robust solutions for identifying rare pathogenic variants.
• The models enable enhanced genotype-phenotype association studies.
• Accessible resources (scores and software) are provided for broader research adoption.