What a call means

SigMA2 makes a signature call, an HRD call and an MMRD or POLE call, and each is a prediction from a different reading of the tumor's mutations. This page states, for each, what the classifier reads, how it was trained, what a positive call supports, and what it does not establish.

Three readings of one tumor

A signature call

A mutational signature is a pattern of substitution types in their trinucleotide context, extracted from cancer genomes and exomes by decomposition (Alexandrov et al., 2013; 2020). Signature 3 is associated with biallelic inactivation of BRCA1 or BRCA2, with BRCA1 or RAD51C silenced by promoter methylation, and with germline truncating PALB2 variants (Polak et al., 2017); it is also found in tumors without a BRCA1/2 mutation (Nik-Zainal et al., 2016).

A signature call from SigMA2 says that the panel's SNV spectrum resembles the tumor type's genomes that carry the signature more than its genomes that do not, at a false-positive rate held on simulated panels, by the method of SigMA (Gulhan et al., 2019; how a classifier scores a panel). The classifier reads the 96-channel spectrum and the SNV count; it reads no indel, no copy number and no gene.

What it supports: the tumor's point mutations carry the signature's pattern at a strength the panel can show. What it does not establish: the cause. A Signature 3 call is not a BRCA1/2 test, and it is not a determination of homologous-recombination deficiency. The signature records mutational processes that acted over the tumor's history: repair can be restored, by a reversion mutation for instance (Waks et al., 2020), and recommendations of a working group of the European Society for Medical Oncology, on ovarian cancer, describe mutational signatures and genomic scars as associated with a history of HRD and ask for better tests of current homologous-recombination status (Miller et al., 2020). The breast Signature 3 classifiers on msk and tcga_mc3 were trained on strong Signature 3 only, and a weak signature is expected to be negative (known limits).

A negative call is weak evidence of absence. The cutoffs hold the false-positive rate, not the sensitivity: on a panel, a tumor whose few mutations include few of the signature's is missed (when a call is possible), so a count of calls is not the signature's prevalence in a cohort. A signature run does not check for hypermutation either: a Signature 3 call on a tumor the MMRD/POLE classifier calls MMRD or POLE, or on a panel dominated by APOBEC mutations, is read off a spectrum that other processes shaped, and deserves a look at the spectrum before it is reported.

An HRD call

Homologous-recombination deficiency leaves several records in a genome: Signature 3 in the point mutations, deletions with microhomology at the breakpoints (Alexandrov et al., 2020), and copy-number scars: loss of heterozygosity (Abkevich et al., 2012), telomeric allelic imbalance (Birkbak et al., 2012) and large-scale state transitions (Popova et al., 2012), whose sum is the HRD score of Telli et al. (2016). HRDetect weighted six features from substitution, insertion-deletion and rearrangement patterns into one predictor of BRCA1/2 deficiency on whole genomes (Davies et al., 2017).

SigMA2's HRD model is a gradient-boosting classifier per platform over three groups of features: SigMA v1's Signature 3 score, likelihood and exposure for the tumor; the counts of deletions of at least five bases with and without a microhomology of at least two bases; and, when copy-number segments are given, either the three scars from allele-specific segments or the fraction of genome altered and a log-ratio transition count from panel segments. It was trained in two steps on tumors with biallelic BRCA1/2 inactivation plus, for each, the three most similar BRCA1/2-wild-type tumors, relabeled once by out-of-fold scores (the preprint of Gulhan et al., 2025, which describes the TCGA exome model; the panel models apply the procedure to other cohorts with the truth each allows, listed in what each HRD model was trained on); mismatch-repair deficient and POLE tumors were left out of training. Where SigMA v1 has no Signature 3 classifier for the tumor type and platform, the model reads no Signature 3 score at all. Two models score every tumor, a pan-cancer one and the tumor type's own or a pooled one; HRD means both score above 0.5, HRP neither, and HRD-low means the two disagree.

What it supports: the tumor's mutations and copy number resemble those of BRCA1/2-deficient tumors of the training cohorts, more than those of the rest. What it does not establish: the current state of homologous recombination in the tumor. Every feature it reads is a record of the tumor's history, repair can be restored after the scars were made (Waks et al., 2020), and the ESMO working group asks for better tests of current repair status (Miller et al., 2020). A tumor below the SNV floor is called on its copy number and deletion counts alone, or, without segments, on no feature of its own; its HRP is not an assessment. The call also depends on what was given: without segments the model runs without copy number, and a panel's few SNVs give the Signature 3 features little to read (when a call is possible).

A Signature 3 call and an HRD call are two answers to two questions. Report them side by side and do not substitute one for the other: a tumor can carry Signature 3 and be called HRP. Two Signature 3 classifiers can also disagree: the HRD model reads SigMA v1's, and a soi="SBS3" run uses SigMA2's classifier for the cell (worked example: HRD on exomes shows a tumor on which they do).

An MMRD or POLE call

Defective mismatch repair is associated with the substitution signatures SBS6, SBS15, SBS21, SBS26 and SBS44, and, with a concurrent POLE or POLD1 proofreading defect, SBS14 and SBS20 (Alexandrov et al., 2020); it also leaves small insertions and deletions at microsatellites, the basis of microsatellite-instability testing (Luchini et al., 2019) and of the sequencing-based callers MSIsensor (Niu et al., 2014; Middha et al., 2017) and MANTIS (Kautto et al., 2017). Mutations in the POLE exonuclease domain produce SBS10a and SBS10b, usually with more than 100 mutations per megabase (COSMIC, SBS10a).

SigMA2's MMRD/POLE classifier is a three-class gradient-boosting model per platform over the likelihoods of SigMA v1's MSI and POLE reference clusters, the likelihood ratios and exposures of mismatch-repair and POLE signatures, and the counts of insertions and deletions at the bundled repeat loci; the msk, fo and op models also read cosine similarities, the SNV count and all deletions, the tcga_mc3 and seqcap models all insertions. The shipped models were trained on three classes. On MSK-IMPACT panels, MMRD was an MSIsensor call of instability and POLE a pathogenic POLE exonuclease-domain hotspot (León-Castillo et al., 2020) in a microsatellite-stable tumor. On TCGA exomes (the MC3 calls, Ellrott et al., 2018), MMRD needed MANTIS above 0.4 (Bonneville et al., 2017) and MSIsensor at 10 or more, MMRP both callers clearly stable, and tumors in between or on which they disagreed were left out; on MSK-IMPACT, unstable tumors with few microsatellite indels were left out of training, so borderline instability was not trained; the exome-derived panel models were trained on those exomes reduced to each panel's footprint. MMRD_category is the most probable of the three classes; a per-class cutoff set at a 1% false-positive rate in cross-validation is also required, but at its values it never overrules the most probable class.

What it supports: the tumor's spectrum and microsatellite indels resemble those of tumors a sequencing-based MSI caller called unstable. What it does not establish: loss of a mismatch-repair protein, which immunohistochemistry of MLH1, MSH2, MSH6 and PMS2 determines, or the result of a PCR or NGS microsatellite-instability test on the tumor itself (Luchini et al., 2019). The training truth was itself a sequencing-based call. The cutoffs hold the false-positive rate at 1%, so a positive call is the stronger statement: a panel with few mutations and few indels at microsatellites gives the classifier little to read, and an MMRP call there says little. A mismatch-repair and a POLE defect can coincide (Alexandrov et al., 2020), and a tumor with both was labeled MMRD in training, while León-Castillo et al. (2020) found that endometrial carcinomas with both behave like POLE-ultramutated tumors: an MMRD call does not exclude a pathogenic POLE mutation.

Before a call goes into a report

• Name the call and the classifier: SBS3, HRD or MMRD, the release, the platform, the tumor type and the operating point (5. Read the output).
• Keep the Signature 3 call and the HRD call apart, and report both.
• State what was not read: no copy number when no segments were given, no indels when the input was a matrix, no microsatellite counts when nmsi_ins and nmsi_del are missing from the output.
• Leave tumors below the SNV floor out of an HRD prevalence, and do not read a count of calls as a signature's prevalence: a negative call on a panel is weak evidence of absence.
• A clinical determination of HRD or of MMR deficiency rests on an assay made for it; these calls rank and flag tumors for one.

References

• Abkevich V, et al. Patterns of genomic loss of heterozygosity predict homologous recombination repair defects in epithelial ovarian cancer. Br J Cancer 2012. 10.1038/bjc.2012.451
• Alexandrov LB, et al. Signatures of mutational processes in human cancer. Nature 2013. 10.1038/nature12477
• Alexandrov LB, et al. The repertoire of mutational signatures in human cancer. Nature 2020. 10.1038/s41586-020-1943-3
• Birkbak NJ, et al. Telomeric allelic imbalance indicates defective DNA repair and sensitivity to DNA-damaging agents. Cancer Discov 2012. 10.1158/2159-8290.CD-11-0206
• Bonneville R, et al. Landscape of microsatellite instability across 39 cancer types. JCO Precis Oncol 2017. 10.1200/PO.17.00073
• Davies H, et al. HRDetect is a predictor of BRCA1 and BRCA2 deficiency based on mutational signatures. Nat Med 2017. 10.1038/nm.4292
• Ellrott K, et al. Scalable open science approach for mutation calling of tumor exomes using multiple genomic pipelines. Cell Syst 2018. 10.1016/j.cels.2018.03.002
• Gulhan DC, et al. Detecting the mutational signature of homologous recombination deficiency in clinical samples. Nat Genet 2019. 10.1038/s41588-019-0390-2
• Gulhan DC, et al. Tissue specificity and chromosomal alterations shape divergent immune programs in HRD tumors. bioRxiv 2025, preprint, not peer-reviewed. 10.1101/2025.10.13.681986
• Kautto EA, et al. Performance evaluation for rapid detection of pan-cancer microsatellite instability with MANTIS. Oncotarget 2017. 10.18632/oncotarget.13918
• León-Castillo A, et al. Interpretation of somatic POLE mutations in endometrial carcinoma. J Pathol 2020. 10.1002/path.5372
• Luchini C, et al. ESMO recommendations on microsatellite instability testing for immunotherapy in cancer, and its relationship with PD-1/PD-L1 expression… Ann Oncol 2019. 10.1093/annonc/mdz116
• Miller RE, et al. ESMO recommendations on predictive biomarker testing for homologous recombination deficiency and PARP inhibitor benefit in ovarian cancer. Ann Oncol 2020. 10.1016/j.annonc.2020.08.2102
• Middha S, et al. Reliable pan-cancer microsatellite instability assessment by using targeted next-generation sequencing data. JCO Precis Oncol 2017. 10.1200/PO.17.00084
• Nik-Zainal S, et al. Landscape of somatic mutations in 560 breast cancer whole-genome sequences. Nature 2016. 10.1038/nature17676
• Niu B, et al. MSIsensor: microsatellite instability detection using paired tumor-normal sequence data. Bioinformatics 2014. 10.1093/bioinformatics/btt755
• Polak P, et al. A mutational signature reveals alterations underlying deficient homologous recombination repair in breast cancer. Nat Genet 2017. 10.1038/ng.3934
• Popova T, et al. Ploidy and large-scale genomic instability consistently identify basal-like breast carcinomas with BRCA1/2 inactivation. Cancer Res 2012. 10.1158/0008-5472.CAN-12-1470
• Telli ML, et al. Homologous recombination deficiency (HRD) score predicts response to platinum-containing neoadjuvant chemotherapy in patients with triple-negative breast cancer. Clin Cancer Res 2016. 10.1158/1078-0432.CCR-15-2477
• COSMIC Mutational Signatures, SBS10a and SBS10b, proposed etiology. cancer.sanger.ac.uk/signatures/sbs/sbs10a, accessed 2026-11-05.
• Waks AG, et al. Reversion and non-reversion mechanisms of resistance to PARP inhibitor or platinum chemotherapy in BRCA1/2-mutant metastatic breast cancer. Ann Oncol 2020. 10.1016/j.annonc.2020.02.008