PolyBoost: An enhanced genomic variant classifier using extreme gradient boosting.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Human exome sequencing yields numerous variants with unknown clinical impact.
• Interpreting these variants of uncertain significance is crucial for diagnosing genetic disorders.
• In silico predictive classifiers are essential tools for variant interpretation.

Purpose of the Study:

• To improve the predictive performance of in silico variant classifiers.
• To investigate alternative machine learning models beyond the naive Bayes approach used by PolyPhen-2.
• To develop a more accurate tool for classifying genetic variants.

Main Methods:

• Retrained PolyPhen-2 classifiers using extreme gradient boosting (XGBoost), random forests, artificial neural networks, and support vector machines.
• Utilized the PolyPhen-2 feature set for retraining.
• Externally validated classifiers on ClinVar pathogenic and benign variants not included in training datasets.

Main Results:

• An XGBoost-based classifier, named PolyBoost (PolyPhen-2 Booster), demonstrated improved discriminative performance and calibration compared to PolyPhen-2.
• External validation on ClinVar data confirmed PolyBoost's superior accuracy.
• The developed software is freely available.

Conclusions:

• PolyBoost offers enhanced interpretation of clinical exome sequences for identifying monogenic diseases.
• It can be integrated into existing bioinformatics workflows as a post-analysis tool.
• This improves the clinical relevance of genetic variant analysis.