ClairS: a deep-learning method for long-read tumor-normal pair somatic small variant calling.

Deep-learning somatic small-variant caller for long-read tumor-normal pairs, leveraging long-read phasing

Somatic variant discovery is crucial in clinical oncology, but most tools are designed for short reads. The authors present ClairS (Clair-Somatic), a deep-learning somatic variant caller dedicated to long-read tumor-normal pairs. Trained on synthetic somatic variants across coverages and allele fractions, it detects a wide range of variants: on the Nanopore HCC1395 dataset at 50/25×, it reaches F1 scores of 89.83% (SNVs) and 73.38% (indels), rising to 96.19% and 79.67% with training augmented by real cell lines. The improved phasing enabled by long reads is key for low-allele-fraction SNVs. ClairS is open source.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Fills a real gap: a robust, open-source somatic caller designed for long reads, whereas the ecosystem remained short-read dominated. As long-read gains ground in oncology (SVs, phasing), this is a useful infrastructure component, whose performance on real tumors beyond cell lines will need confirmation.

Analysis by Dr Thibaut Benquey