Comparison of missing data handling methods for variant pathogenicity predictors.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Clinical genetic testing relies on next-generation sequencing (NGS) to identify disease-causing variants.
• Manual variant evaluation is impractical for the high volume and speed required in clinical settings.
• Machine learning and AI tools for variant pathogenicity prediction face challenges due to sparse genetic data.

Purpose of the Study:

• To introduce AMISS, an open-source framework for evaluating methods of handling missing genetic variant data.
• To assess the impact of different missing data imputation strategies on variant pathogenicity prediction.
• To identify optimal methods for managing missing data in clinical genetic variant analysis.

Main Methods:

• Development and application of the AMISS framework.
• Evaluation of 14 distinct methods for handling missing genetic variant data.
• Comparative analysis of imputation methods based on precision, computational cost, and other performance metrics.

Main Results:

• Significant performance variation observed across the 14 evaluated missing data handling methods.
• Simpler imputation techniques, particularly mean imputation, demonstrated superior performance.
• Method selection impacts accuracy, reliability, speed, and computational expenses in variant pathogenicity prediction.

Conclusions:

• Effective management of missing genetic variant data is crucial for accurate and efficient clinical genetic testing.
• Mean imputation offers a robust and performant solution for handling missing data in variant pathogenicity prediction.
• The AMISS framework facilitates the systematic evaluation and selection of appropriate methods for missing data imputation.