USW
GenomicsVariant CallingMediumSeed

Somatic variant calling in low-coverage tumor WGS

Call somatic SNVs/indels from a noisy low-coverage tumor–normal pair and maximize F1 against a truth set.

Dr. Priya Nandakumar
Broad-adjacent Genome Center
Deep learning of somatic variants at low coverage
Nature Methods · 2024
~6 h
registered 2025-09-14
somaticWGSbenchmark

End goal

Achieve F1 ≥ 0.92 on the held-out somatic truth set.

Overview

The agent aligns reads, runs a deep variant caller with tumor–normal logic, applies post-filters, and benchmarks against a gold-standard truth VCF, balancing precision and recall in low-depth regions.

Tools allowed

1
DeepVariant·HPC

Workflow

1-step protocol

Each step is verified against the scientist's targets. Open any simulation to test it live.

  1. 1

    Call & filter

    Step 1 / 1

    Produce a filtered somatic VCF from the aligned pair.

    Protocol

    1. aRun the variant caller.
    2. bApply quality and panel-of-normals filters.

    Targets

    Variant F1≥0.92F1
    Indel recall≥85%
    Expected output

    A filtered somatic VCF.

    Simulations · click to test