Onko-Gene

Critical literature watch in the molecular oncology of solid tumors

FREN

Bioinfo & AI

SomatiX: accurate and efficient somatic SNV detection from matched tumor-normal long-read DNA sequencing

Summary

SomatiX is a deep-learning tool designed for accurate somatic single-nucleotide variant (SNV) calling from matched tumor-normal long-read sequencing data across ONT and PacBio platforms. The method utilizes an allele-aware pileup representation to overcome sequencing noise, variable tumor purity, and low variant allele fractions (VAFs). Benchmarked across cancer cell line datasets, SomatiX outperformed DeepSomatic and ClairS in F1 scores on Oxford Nanopore data and matched performance on PacBio, reaching mean F1 scores of 0.986 and 0.988 on held-out chromosome 1 (for VAF ≥ 0.05 and ≥3 supporting reads). In addition to robust detection of low-VAF SNVs, it features a compact model architecture and the fastest runtime among evaluated tools.

Critical appraisal

SomatiX introduces a compact deep-learning architecture utilizing an allele-aware pileup representation to effectively distinguish long-read sequencing errors from true somatic SNVs. It was benchmarked against leading state-of-the-art tools, DeepSomatic and ClairS, across cancer cell-line datasets and varied sequencing depths, outperforming them on Oxford Nanopore data (mean F1 of 0.986) and showing competitive accuracy on PacBio (mean F1 of 0.988) at VAF ≥ 0.05. The tool is fully open-source with pre-trained models and scripts made available on GitHub and Zenodo, supporting transparent reproducibility. Clinically, its reduced computational footprint and superior sensitivity for low-VAF variants streamline the implementation of tumor-normal long-read workflows in translational and molecular tumor board pipelines.

The summary and the appraisal were written by a language model from the title and abstract of the original article. The article was selected automatically by a model, without human review. Methodology