Empirical Bayes single nucleotide variant-calling for next-generation sequencing data.

Area of Science:

• Computational genomics
• Bioinformatics
• Cancer research

Background:

• Accurate identification of single nucleotide variants (SNVs) is crucial for cancer genomics.
• Existing SNV calling algorithms exhibit significant disagreement on real-world datasets.
• There is a need for robust and reliable methods for germline SNV detection and prioritization.

Purpose of the Study:

• To develop a novel local false discovery rate (LFDR) estimator for germline SNV calling.
• To create an LFDR-based algorithm for prioritizing SNVs called by other variant-calling tools.
• To evaluate the performance of the proposed LFDR approach against existing state-of-the-art methods.

Main Methods:

• An empirical Bayesian approach was used to develop the LFDR estimator.
• The method learns model parameters without prior information, utilizing genomic region-wide data.
• A second LFDR-based algorithm was developed for prioritizing variant calls from other algorithms.

Main Results:

• The proposed LFDR approach demonstrated performance comparable to or exceeding widely used SNV calling programs on gold-standard cell line data.
• Prioritizing variant calls using the LFDR score allowed for significant increases in precision with minimal loss of sensitivity.
• The LFDR method effectively addresses the discrepancies observed among different state-of-the-art SNV callers.

Conclusions:

• The developed LFDR estimator provides a robust and accurate method for germline SNV calling in cancer genomics.
• The LFDR-based prioritization algorithm offers a valuable tool for refining variant calls and improving precision.
• This approach enhances the reliability of SNV identification, crucial for advancing cancer research and diagnostics.