Analysis of machine learning algorithms as integrative tools for validation of next generation sequencing data

Area of Science:

• Genomics
• Bioinformatics
• Clinical Diagnostics

Background:

• Next-generation sequencing (NGS) is the preferred technology for clinical diagnostics.
• Sanger sequencing is commonly used for orthogonal confirmation of NGS results, often proving redundant for high-quality data.
• Reducing redundant confirmations can streamline diagnostic workflows.

Purpose of the Study:

• To establish criteria for distinguishing NGS variant calls that require orthogonal confirmation.
• To assess the feasibility of using machine learning (ML) for this purpose.
• To decrease the workload in genetic diagnostic laboratories.

Main Methods:

• Trained and tested various ML algorithms on a dataset of 7976 NGS calls previously confirmed by Sanger sequencing.
• Varied training dataset size and class balance to evaluate ML performance.
• Determined conditions for ML validity in a clinical diagnostic setting.

Main Results:

• ML is a viable approach for identifying NGS variant calls needing further investigation.
• High accuracy in clinical settings requires sufficient and well-balanced training data (true/false positive NGS calls).

Conclusions:

• Integrating ML into the NGS validation workflow can reduce the number of Sanger confirmations needed.
• This integration streamlines the diagnostic process, decreasing turnaround time and costs for high-quality NGS calls.