Gene-specific pathogenicity predictor for chromatin remodeling BAF complex-associated neurodevelopmental disorders.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Whole genome sequencing identifies numerous variants of uncertain significance (VUS), creating diagnostic challenges.
• Genetic counselors require accurate tools to differentiate pathogenic variants from VUS for patient diagnosis.
• Existing machine learning (ML) tools often lack the gene-specific accuracy needed for clinical application.

Purpose of the Study:

• To develop a workflow for creating accurate, gene-specific, ensemble-learning ML models for predicting variant pathogenicity.
• To improve the classification of variants of uncertain significance (VUS) in rare neurodevelopmental disease genes.

Main Methods:

• Leveraged outputs from multiple ML algorithms, variant locations, and evolutionary conservation data.
• Screened 15 ML algorithms using variants in SMARCA2 and SMARCA4 associated with neurodevelopmental diseases.
• Tuned a random forest learner and developed an ensemble predictor for BAF complex proteins.

Main Results:

• Achieved 0.93 accuracy with a tuned random forest learner on holdout data for specific genes.
• Developed an ensemble predictor for BAF complex proteins achieving 0.91 accuracy.
• The BAF complex-specific predictor outperformed existing methods in accuracy and AUPRC.

Conclusions:

• Gene-specific ML model calibration is crucial for accurate variant pathogenicity prediction.
• The developed workflow offers an efficient and cost-effective method for enhancing ML tools for genetic counselors.
• Improved ML tools can alleviate diagnostic bottlenecks and aid in identifying causative variants for rare diseases.