Somatic variant-calling beyond cancer: Repurposing algorithms to map low‑allele‑fraction variants across genomics.

BACKGROUND

Somatic variant callers were originally developed to identify tumour-specific mutations in mixed tumour-normal samples. Increasingly, disciplines such as developmental biology, reproductive medicine, virology and mitochondrial genetics require detection of low-frequency variants from high-depth sequencing. Many laboratories therefore reuse cancer callers without clear guidance on their statistical assumptions or validation in non-cancer contexts.

METHODS

We review widely used somatic callers and post-calling classifiers, summarising their underlying models and assumptions. We then synthesise peer-reviewed case studies (2020-2025) where these tools were applied to detect low-allele-fraction variants outside cancer. For each study, we extract sample type, sequencing depth, variant-allele-fraction thresholds, caller(s) used and validation approaches. We discuss technical challenges and propose practical adaptations.

RESULTS

Cancer callers generally assume diploid genomes and moderate allele fractions, use Beta-binomial or Poisson-based models and apply filters for tumour/normal comparisons. In non-cancer applications, allele fractions frequently fall below 5%, confounded by ploidy differences and sequencing artefacts. Published studies demonstrate that somatic callers can be repurposed for detecting post-zygotic variants at 1-3% allele fraction, sperm mosaicism, mitochondrial heteroplasmy as low as 0.4-0.5%, minority viral variants around 5% and cfDNA variants at ≥ 1-5% allele fraction. Unique molecular identifiers, panel-of-normal filtering and machine-learning post-processing markedly improve specificity. Emerging long-read and deep-learning callers (e.g. DeepSomatic) promise enhanced sensitivity.

CONCLUSIONS

Somatic callers offer versatile frameworks to interrogate low-allele-fraction variants across genomics; however, careful parameter tuning and validation are essential. Future work could integrate pangenome references and federated benchmarking datasets to make somatic variant detection more robust in diverse biological settings.