Cancer Mutation Atlas CMA

Karchin Lab · Johns Hopkins University

Cancer Mutation Atlas

How often is a mutation found in a particular cancer type? The Cancer Mutation Atlas brings together somatic cancer sequencing data from multiple studies to estimate mutation prevalence across patients and cancer types.

Developed by the Karchin Lab at Johns Hopkins University, CMA helps researchers and clinicians put observed variants in context. Cancer diagnoses are organized with OncoTree, making it possible to explore a specific diagnosis or a broader group of related cancers.

Use the atlas through open data downloads, the CMA annotator in OpenCRAVAT, or the web interface.

Atlas information

Counting mechanics

Patients are the counting unit. Multiple samples from the same patient and diagnosis are combined. A mutation observed in any of those samples counts once, and the patient’s tested regions are the union of their samples’ sequencing panels.

Prevalence at a genomic position

Patients carrying the variant ÷ patients tested at that position

The denominator changes by position: a patient contributes only where their assay provides coverage. Coverage is based on sequencing designs, so it describes the regions assayed rather than guaranteeing a particular read depth in every patient.

Both tumor-only and matched tumor–normal studies contribute. Some studies compare a tumor with a normal sample from the same patient; others sequence the tumor alone. CMA applies study-specific quality checks and a shared germline filter to reduce inherited variants and technical artifacts.

OncoTree connects specific diagnoses to broader cancer groups. Each patient is assigned a cancer code. The atlas stores counts at those assigned codes; the OpenCRAVAT annotator and web widget combine a group’s direct counts with its descendants. For example, lung adenocarcinoma contributes to non-small cell lung cancer and lung totals. Both carrier and tested-patient counts propagate upward, and prevalence is recalculated from those totals.

The precise unit is patient × OncoTree diagnosis. A patient with two distinct diagnoses contributes to each; both units can contribute to a shared parent group. Broader totals therefore need not represent distinct biological patients.

Filtering

Study preparation removes calls or samples that fail the source study’s quality criteria or CMA’s study-specific artifact checks. Variants are standardized to GRCh38 (hg38), checked against the reference genome, and restricted to the regions assayed. Read-support filters apply where supported by the study’s data and policy.

Germline filtering reduces inherited variation in the final atlas. After studies are combined, CMA applies three checks:

  1. Population frequency. A variant is flagged if its allele frequency exceeds 0.04% in any of six population resources: gnomAD v4, ALFA, Regeneron, All of Us, ESP6500, or 1000 Genomes. These frequencies are annotated with OpenCRAVAT.
  2. Panels of normals. Variants found in the GATK panel of normals are flagged; a GDC panel can also be included when configured.
  3. Cancer hotspot rescue. An allele listed in Cancer Hotspots v2 is retained even if one of the shared checks flags it, preserving known recurrent somatic events.

Source-level filters still apply before this shared step. For example, Foundation Medicine calls flagged by its own germline or panel-of-normals assessment are removed during preparation. The QC report records attrition and the effects of filtering for each release.

Studies and counts

CMA brings together targeted panels, exomes, and whole genomes, spanning broad cancer cohorts and studies focused on particular diagnoses.

Study coverage in the CMA workflow
StudiesFocus
AACR Project GENIE, TCGA, Foundation Medicine, CPTACMultiple cancer types, using clinical panels or exome sequencing
OrigiMedPan-cancer targeted sequencing in a Chinese cohort
REBC, ALCHEMISTPapillary thyroid cancer and lung cancer, respectively
Prostate 1000, METABRICProstate and breast cancer, respectively
IWG-MDS, Broad CLL, Duke DLBCLMyelodysplastic syndromes, chronic lymphocytic leukemia, and diffuse large B-cell lymphoma

Release sample and mutation counts: see the beta1 QC report for the included studies, input samples, retained patient units, mutation counts, and filtering losses. Sample counts describe study preparation; prevalence uses the patient units retained after collation.

Version history

beta1 Beta release

Initial beta of the Cancer Mutation Atlas: patient-level somatic mutation counts on GRCh38, OncoTree cancer grouping, and germline-filtered data.

Read the beta1 QC report

Using the CMA

Download the data

Use the germline-filtered atlas for your own analyses. The TSV contains one row per variant with cancer-type and study counts. Parquet files provide one row per variant and cancer type or study, ready for Python, R, or SQL.

Coordinates use GRCh38 and 1-based VCF positions. Downloads retain directly assigned OncoTree counts; combine counts with the matching OncoTree hierarchy when analyzing broader groups. Record the atlas version with your results.

Annotate variants with OpenCRAVAT

The CMA annotator adds mutation counts and tested-patient denominators by cancer type to a variant analysis. Its viewer widget presents prevalence across the OncoTree hierarchy.

Check the reported match type: an exact allele match, a nearby protein-change match, and a genomic-location match describe different sets of atlas observations.

CMA annotator instructions

Explore on the web

Explore how mutation prevalence varies between cancer types using the CMA locus browser. Expand the cancer tree to compare related diagnoses and inspect carrier counts alongside the number of patients tested.

Open the locus browser

From a published release this opens the atlas browser for the latest version; the copy bundled in the repo uses example variants and demonstration counts.