A machine learning model to determine the accuracy of variant calls in capture-based next generation sequencing

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Next-generation sequencing (NGS) is a cornerstone of clinical genetic testing.
• NGS variant call quality is influenced by various factors, necessitating orthogonal confirmation for a subset of calls.
• Orthogonal technologies like Sanger sequencing are currently used to validate NGS results.

Purpose of the Study:

• To develop a deterministic machine-learning model to differentiate high-confidence NGS variant calls from low-confidence ones.
• To reduce the reliance on orthogonal confirmation for NGS results in clinical settings.
• To improve the reliability and efficiency of clinical genetic testing.

Main Methods:

• A machine-learning model was developed using a dataset of 7179 variants.
• The model incorporated sequence characteristics and variant call quality signals.
• The model was trained to minimize false positives, defined as variants called by NGS but not confirmed by Sanger sequencing.

Main Results:

• The model achieved 99.4% accuracy in differentiating variant call confidence.
• 92.2% of variants were classified as high confidence, with 100% confirmed by Sanger sequencing.
• Among low-confidence variants, 92.1% were identified as artifacts by Sanger sequencing.

Conclusions:

• NGS data possesses sufficient features for a machine-learning model to reliably distinguish low from high confidence variants.
• The model effectively identifies variants requiring orthogonal confirmation, enhancing clinical utility.
• Incorporating site-specific and variant call features is crucial for model performance.