PolyBoost

PolyBoost applies extreme gradient boosting (XGBoost) to reclassify PolyPhen-2 outputs, improving discriminative performance and calibration for genomic variant interpretation in clinical exome sequences.

Key Features:

• Post-analysis processing: Accepts and reprocesses outputs produced by PolyPhen-2 to generate revised variant classifications.
• Primary algorithm: Replaces PolyPhen-2's naïve Bayes classifier with XGBoost for improved discrimination and calibration.
• Feature set: Utilizes the feature set derived from PolyPhen-2 as input to the machine learning models.
• Alternative methods evaluated: Employs or considers additional machine learning techniques including random forests, artificial neural networks, and support vector machines.
• External validation: Evaluated using ClinVar datasets containing "pathogenic" and "benign" variants that were excluded from training.
• Target data scope: Designed for interpretation of human exome variant collections on the order of 15,000–20,000 variants, including variants of uncertain significance.

Scientific Applications:

• Clinical exome variant classification: Improves classification accuracy for variants detected in clinical exome sequencing.
• Variant of uncertain significance prioritization: Aids interpretation and prioritization of variants labeled as uncertain in clinical sequencing data.
• Monogenic disease detection: Supports identification and prioritization of candidate variants for monogenic diseases.
• Benchmarking predictive performance: Enables comparative evaluation against PolyPhen-2 using ClinVar pathogenic and benign variants.

Methodology:

Processes PolyPhen-2 outputs and trains an XGBoost classifier that replaces PolyPhen-2's naïve Bayes model using PolyPhen-2 features, with comparisons or consideration of random forests, artificial neural networks, and support vector machines, and external validation on ClinVar pathogenic and benign variants excluded from training.