New method reads sequencing errors directly from overlapping reads to spot rare mutations

Detecting rare somatic mutations in DNA sequencing data is a persistent challenge: a true variant carried by only a handful of DNA molecules can be nearly indistinguishable from the noise created by sequencing errors. A new method called Better Base Quality, or BBQ, tackles this problem by learning a sample’s own error patterns directly from its data, and it reports higher precision in calling low-frequency variants than several widely used tools.

The method exploits a feature of standard paired-end sequencing that is often overlooked. When short DNA fragments are sequenced from both ends, the two reads typically overlap in the middle. If the two reads disagree about the base at a position, that discordance is direct evidence of a sequencing error in one of the reads. Rather than simply discarding such discordant bases, BBQ uses them to build a model of how errors behave in a given sample.

Specifically, BBQ estimates sample-specific error rates conditional on the mutation type, the sequence context surrounding the base, and the base quality score assigned by the sequencer. The model also accounts for two additional effects: the reduced error rate observed for bases that are concordant across overlapping read pairs, and position-specific evidence of sequencing errors. The resulting error probabilities are fed into a likelihood-based variant caller that is designed to prioritize precision when variants are supported by only a few DNA fragments.

Notably, the approach requires neither matched training samples nor molecular barcodes, which are commonly used in ultra-sensitive mutation detection but add cost and experimental complexity. Because BBQ works from standard sequencing data, it can in principle be applied to datasets that were not generated with rare-variant detection in mind.

The researchers evaluated BBQ using a synthetic low-frequency somatic mutation benchmark derived from a human testis sample, together with two independent cell-free DNA datasets from colorectal cancer patients.

Across these benchmarks, the estimated error models captured substantial differences in error patterns between samples, and they improved the discrimination between true alternative alleles and sequencing errors beyond what standard base quality scores alone could provide.

In head-to-head comparisons, BBQ achieved higher precision than commonly used somatic variant callers, including Mutect2, Strelka2, and LoFreq, with the largest gains seen for variants supported by few alternative alleles — precisely the cases where false positives are hardest to avoid. The performance improvements also generalized to an independent cell-free DNA sample that was not part of the model training, suggesting the method is not simply overfitting to the datasets used to develop it.

The broader implication is that overlapping paired-end reads contain more usable information than is typically extracted from them. As the authors note, these overlaps can be used not only to remove or flag discordant bases, but also to learn the specific error patterns of a sample. Incorporating those learned probabilities directly into the variant calling step turns what is usually treated as noise into a signal about the reliability of every base call.

According to the authors, BBQ may be particularly useful for applications in which rare variants must be distinguished from sequencing errors while maintaining a low false-positive rate, as it improves precision without requiring new laboratory workflows. The work is published as an open access article in BMC Biology, with a version of record to follow.

Subject of Research: A computational method that models sample-specific sequencing error rates from overlapping paired-end reads to improve detection of rare somatic mutations.

Article Title: Accurate calling of low-frequency somatic mutations by sample-specific modeling of error rates

Article References: Lin, Y., Lääs, C., Porsborg, P. S., Frydendahl, A., Rasmussen, M. H., Bataillon, T., Schierup, M. H., Andersen, C. L., Almstrup, K., & Besenbacher, S. (2026). Accurate calling of low-frequency somatic mutations by sample-specific modeling of error rates. BMC Biology. https://doi.org/10.1186/s12915-026-02754-w

Image Credits: AI Generated

DOI: 10.1186/s12915-026-02754-w

Keywords: somatic mutations, variant calling, sequencing errors, paired-end sequencing, cell-free DNA, BBQ, bioinformatics, precision, low-frequency variants, next-generation sequencing, colorectal cancer, error modeling