Reducing Sanger confirmation testing through false positive prediction algorithms.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Clinical genome sequencing (cGS) is standard practice, but orthogonal confirmatory testing increases turnaround time and cost.
• Reducing confirmatory testing is crucial for efficient clinical genomic workflows.

Purpose of the Study:

• To evaluate machine learning models for identifying false positive variants in cGS data.
• To decrease the reliance on orthogonal confirmatory testing in clinical genomics.

Main Methods:

• Sequenced five reference human genome samples from the Genome in a Bottle Consortium (GIAB).
• Trained machine learning models to identify false positive variants against established truth sets.
• Compared model performance in detecting single-nucleotide variants (SNVs) and insertions/deletions (indels).

Main Results:

• Models identified 99.5% of false positive heterozygous SNVs and indels.
• Reduced confirmatory testing for nonactionable SNVs by 85% and indels by 75%.
• Overall orthogonal testing (Sanger sequencing) decreased by 71% in clinical practice.

Conclusions:

• Machine learning effectively maintains a low false positive rate in cGS.
• Significant reduction in confirmatory testing is achievable, improving efficiency and cost-effectiveness.
• The developed framework is publicly available for broader application.