SomaticCombiner: improving the performance of somatic variant calling based on evaluation tests and a consensus...

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• Accurate identification of somatic variants is crucial for cancer research and clinical applications.
• Tumor heterogeneity, sub-clonality, and sequencing errors present significant challenges in somatic variant detection.
• Existing somatic variant callers have limitations in sensitivity and specificity.

Purpose of the Study:

• To evaluate the performance of multiple somatic variant callers and ensemble methods.
• To develop and validate a robust computational tool for combining variant caller results.
• To improve the detection of low-frequency somatic variants.

Main Methods:

• Performance evaluation of eight primary somatic variant callers and ensemble methods.
• Utilized real and synthetic whole-genome sequencing, whole-exome sequencing, and deep targeted sequencing datasets.
• Developed SomaticCombiner, a software package employing a variant allelic frequency (VAF) adaptive majority voting approach.

Main Results:

• A simple consensus approach significantly improved somatic variant caller performance.
• Ensemble methods based on consensus were more robust and stable than machine learning approaches.
• SomaticCombiner effectively combined multiple callers and maintained sensitive detection of low VAF variants.

Conclusions:

• Consensus-based ensemble methods offer a superior strategy for somatic variant detection compared to individual callers or ML-based ensembles.
• SomaticCombiner provides a valuable tool for enhancing the accuracy and sensitivity of somatic variant identification.
• Improved somatic variant detection has implications for personalized cancer therapy and diagnostics.