Mutational signatures

In one sentence

Mutational signatures are recurring patterns across many DNA changes that can provide evidence about the processes that generated a tumor’s mutations.

The intuition

Suppose a manuscript contains repeated kinds of typing errors. The overall pattern may suggest a sticking key or a particular editing process. Mutational signatures do something similar for deoxyribonucleic acid (DNA): they summarize patterns across many changes. The analogy cannot identify a culprit with certainty, because several processes can leave similar errors and act at different times.

How it works

Sequencing and variant calling first identify DNA changes. Researchers group those changes into categories and analyze their distribution. A single-base substitution (SBS) changes one base; doublet-base substitutions (DBS) change two adjacent bases; insertions and deletions, or indels, add or remove sequence. The mutation and its surrounding sequence can help define a category.

A signature describes a distribution across categories. A tumor may contain a mixture of signatures, so an algorithm estimates the contribution of each. That assignment is a model-based inference, not a label directly attached to every mutation by the sequencing machine. Large primary cancer-genome studies established catalogs of recurring patterns and investigated their causes. Some have strong biological explanations; others remain uncertain. Alexandrov 2013, Alexandrov 2020.

Some patterns are associated with defective DNA repair. SBS3, a particular substitution signature, has been associated with impaired homologous recombination (HR). Association is not a unique diagnosis of a BRCA1 or BRCA2 mutation: a repair process can be disrupted in several ways.

HRDetect is an example of a research classifier combining several genomic features, including substitution, indel and rearrangement patterns, to estimate BRCA-related repair deficiency. It is not simply “SBS3 positive.” Its original training and evaluation used particular tumor datasets and sequencing methods. Davies 2017.

The arrows move from observations to interpretation. They do not move directly from a signature to a drug choice.

Why it matters in cancer

Signatures help researchers investigate exposures, repair defects and tumor evolution. They may reveal a pattern when no single causative gene alteration has been established. That is a reason to investigate further, rather than fill the gap with an assumed mechanism.

A signature summarizes accumulated mutations. It does not directly observe present repair activity. Likewise, a classifier score for a repair state is not the probability that a patient will respond to a drug. That requires separate validation with the relevant endpoint, treatment and population.

Assay card

Field	What to ask
Input and tissue cost	Tumor DNA, often paired with normal DNA; extraction consumes tissue. Accepted specimen quality and amount depend on the method.
How	Call mutations; classify them; fit or extract patterns; assess uncertainty and model fit.
Output and units	Mutation counts assigned to signatures, proportions or estimated contributions; a classifier may instead output a model score. Check its definition.
Thresholds	No universal “signature present” or drug-response cutoff. Record catalog, software version, feature set and validation population.
Failure modes	Too few mutations, artifacts, missed variants, similar signatures competing for assignment and a poor model fit.
Limits and validation	Research inference may not transfer between panels, exomes and whole genomes. A different method needs its own analytical and clinical validation.

In the original HRDetect study, applying whole-genome-derived features to whole-exome data reduced detection performance; retraining changed the result. Missing genomic features are therefore more than an incidental format difference. Davies 2017.

Common confusions

Signature versus a target mutation: a pattern across changes is not one protein altered by one variant.
Amount versus certainty: a small estimated contribution can be unstable when competing explanations fit similarly.
SBS3 versus HRDetect: one feature and a multifeature classifier are not the same test.
Classifier probability versus response probability: the modeled outcome must be named.

Try it

A fictional small panel detects a few mutations. An exploratory fit assigns some to SBS3. Someone copies a whole-genome HRDetect interpretation into the report. What went wrong?

Answer: the data and model are mismatched. Sparse mutations and unavailable rearrangement features can make the estimate unreliable. The correct next step is to check method-specific performance and uncertainty, not infer drug sensitivity.

Explain it back

Why can two tumors share a signature without sharing the same causative mutation?

One possible answer: different disruptions can affect the same process, and some patterns also have competing explanations.

Takeaway

A mutational pattern supports a scoped hypothesis; its meaning depends on the data, model and validation.

Genomic scars focus on chromosome imbalance and rearrangement measures.
Analytical validity, clinical validity and utility separate a reliable computation from useful clinical evidence.

Sources

Source check: 2026-10-09. Signature mechanisms and research interpretation; expert and learner review pending. No research classifier is presented as validated individual treatment sensitivity.