An Ensemble Approach to Predict the Pathogenicity of Synonymous Variants

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Single-nucleotide variants (SNVs) are key genetic variations linked to diseases.
• Synonymous SNVs (sSNVs) affect gene regulation and RNA processing but are difficult to study functionally.
• Existing computational tools primarily focus on non-synonymous SNVs (nsSNVs), neglecting sSNVs.

Purpose of the Study:

• To develop and validate a computational model for predicting the pathogenicity of sSNVs.
• To identify key features influencing the functional impact of sSNVs.
• To reclassify sSNVs with unknown clinical significance.

Main Methods:

• Downloaded sSNVs and associated features (conservation, DNA-RNA, splicing) from ClinVar.
• Applied feature selection to identify informative predictors.
• Developed an ensemble random forest classification algorithm to predict sSNV pathogenicity.

Main Results:

• An ensemble predictor using 20 selected features achieved high accuracy (87%), precision (79%), and recall (91%) in classifying sSNVs.
• The model successfully reclassified sSNVs with unknown clinical significance.
• The developed method demonstrates robustness for predicting the effects of novel sSNVs.

Conclusions:

• The developed random forest model provides an accurate and robust approach for predicting sSNV pathogenicity.
• This tool can aid in functional studies and clinical interpretation of sSNVs.
• The method offers a valuable resource for understanding the genetic basis of disorders caused by sSNVs.