Researchers published DeepSomatic in Nature Biotechnology describing a deep‑learning pipeline for accurate somatic small‑variant discovery across short‑ and long‑read sequencing technologies. The work couples a new variant caller with seven community benchmark datasets (CASTLE‑panel) to evaluate performance across platforms, improving detection of low‑frequency and complex somatic events. The authors also released public resources to enable broader benchmarking and adoption. DeepSomatic aims to harmonize somatic variant calling in cancer genomics and support translational applications such as liquid‑biopsy monitoring and precision oncology trial enrollment.