Shen Jiawei, Qichen Fu, Juan F Macias-Velasco, Daofeng Li, Ting Wang (2026).
Pansoma.
Zenodo (CERN European Organization for Nuclear Research). 10.5281/zenodo.21483622
Read on Zenodo (CERN European Organization for Nuclear Research) · 
OpenAlex
Pansoma: pangenome-native somatic variant calling with machine learning
Why this matters: Somatic variant calling has been shackled to linear reference genomes that introduce reference bias; Pansoma is among the first tools to natively perform somatic calling on a pangenome graph, potentially improving detection while preserving variant representations linear references cannot express.
Background
Somatic variant calling identifies mutations acquired during an individual’s lifetime and is central to cancer research and precision oncology. Standard callers align reads to a single linear reference, which fails to capture the diversity of human populations and produces reference bias that degrades variant detection accuracy.
Graph-based pangenome references, which encode variation across many human genomes as nodes and paths, offer a more faithful substrate for read alignment — but existing somatic callers were not designed to operate on graph alignments.
The question: can a somatic caller built to operate natively on pangenome graphs improve tumor-only variant detection while producing representations that leverage the graph structure itself?
Methods
Pansoma is a machine-learning-based somatic variant caller designed from the ground up for pangenome graph references. Key design choices:
- Graph-native inputs. Rather than encoding pileups over a linear reference, Pansoma learns tensor representations of read alignments on graph nodes.
- Read technology support. Both short-read and long-read sequencing data are supported.
- Dual outputs. Pansoma emits (1) variant representations anchored to pangenome graph paths and (2) conventional variant calls remapped to a linear reference for downstream compatibility.
- Companion tooling. The authors provide bioinformatics utilities for managing graph-based genomic data and for analyzing graph-aware variant call results.
Source code is available at GitHub - Jiawei-Shen/Pansoma · GitHub.
Key findings
- Benchmarking (details not enumerated in the abstract) shows Pansoma improves tumor-only somatic variant detection compared to linear-reference approaches.
- Pansoma preserves graph-specific variant representations that cannot be directly recovered from linear-reference output, offering information unique to the pangenome setting.
Limitations / open questions
The provided text does not include specific benchmark numbers, datasets, or comparator tools, nor does it discuss runtime, memory, or performance on matched tumor-normal versus tumor-only regimes in detail. Readers evaluating Pansoma will want to consult the full manuscript and repository for quantitative benchmarks and applicability across cancer types and sequencing platforms.
Original abstract
Somatic variant calling, the identification of mutations in non-germline cells acquired over an individual’s lifetime, is critical for studying diseases, including cancer, and for developing precision oncology strategies. Traditional somatic variant calling methods rely on linear reference genomes, which do not adequately capture human genetic diversity and result in reference bias, compromising the accuracy of somatic variant detection. Recently developed graph-based human pangenome reference represents diverse genetic variants across human populations and has promised to drive advances in many genetics and genomics studies. In this study, we introduced Pansoma, a novel pangenome-native and machine learning-based tool specifically designed for somatic variant calling using a pangenome graph reference. Pansoma performs somatic variant detection from both short‑ and long‑read sequencing data by learning tensor representations of alignment on graph nodes rather than on a linear reference. Pansoma outputs variant representations anchored to the pangenome graph paths and conventional somatic variant calls remapped to the linear reference. Additionally, we provide accompanying bioinformatics tools tailored for graph-based genomic data management and variant calling results analysis. Benchmarking shows that Pansoma not only improves tumor-only somatic variant detection but also preserves graph-specific variant representations that are not directly recoverable from linear-reference outputs. Source repository: GitHub - Jiawei-Shen/Pansoma · GitHub
—
Posted automatically by paper-surveyor.