Enhancing missense variant classification in predicted intrinsically disordered regions

Area of Science:

• Genomics
• Computational Biology
• Biochemistry

Background:

• Missense variants in intrinsically disordered regions (IDRs) pose a significant challenge for disease classification, as over 25% of deleterious variants occur in these regions.
• Existing in silico predictors perform poorly in IDRs, limiting their clinical utility.

Purpose of the Study:

• To develop and evaluate a machine learning methodology for improved classification of disease-causing missense variants in IDRs.
• To enhance the performance of existing in silico variant predictors by incorporating IDR-specific features.

Main Methods:

• Developed a machine learning model integrating global IDR conformation (gIDRc), phase separation (PS) features, and protein embeddings (ProtTransBertBFD) for wild-type and mutant sequences.
• Defined IDR boundaries using AlphaFold-RSA predictions based on AlphaFold2 pLDDT scores and relative solvent accessibility.
• Utilized ClinVar variant classifications as ground truth for model training and evaluation.

Main Results:

• The baseline model using only IDR-specific features achieved a PR-AUC of 0.817.
• Integrating IDR features significantly improved existing predictors: AlphaMissense-Enhanced (PR-AUC 0.807 to 0.919), ESM1b-Enhanced (PR-AUC 0.679 to 0.845), and EVE (PR-AUC 0.591 to 0.910).

Conclusions:

• The developed methodology effectively enhances the classification of missense variants in IDRs.
• This approach complements existing in silico predictors, offering improved accuracy for variant interpretation in genetic disease research.