Combining multiplexed functional data to improve variant classification.

Area of Science:

• Genomic medicine
• Bioinformatics
• Clinical genetics

Background:

• Variants of uncertain significance (VUS) pose challenges in clinical variant classification.
• Multiplexed assays of variant effect (MAVEs) generate functional data for thousands of genetic variants.
• Multiple MAVEs for a single gene may assess different functional impacts, necessitating data integration.

Purpose of the Study:

• To develop and assess methods for integrating data from multiple MAVEs for improved variant classification.
• To provide guidance on combining multiplexed functional data for clinical interpretation.
• To evaluate the utility of integrated MAVE data in strengthening evidence for variant pathogenicity.

Main Methods:

• Curated published MAVE datasets for TP53, LDLR, and PTEN.
• Applied statistical and machine learning methods (PCA, k-means, Naïve Bayes, random forest) for data integration.
• Assessed integration utility using metrics like sensitivity, specificity, and evidence strength.

Main Results:

• Developed a stepwise process for data curation, model generation, and validation.
• Created a web applet for testing MAVE integration methods and calculating integrated scores.
• Supervised learning methods, particularly random forest, generally outperformed individual MAVE datasets in variant classification.

Conclusions:

• Integrating multiple MAVE datasets strengthens functional evidence for clinical variant classification.
• Researchers can maximize MAVE value by following structured integration processes.
• This approach may reveal novel pathogenicity mechanisms in clinically relevant genes.