Every tumour carries a kind of fingerprint: a set of mutations that, indirectly, tells the story of the biological processes that produced them—from DNA replication errors to faults in repair mechanisms, all the way to specific exposures or treatments.

Building on this idea of a “signature” (mutational signature), a research team at the University of Trieste—made up of PhD students from the Applied Data Science and Artificial Intelligence programme and coordinated by Prof. Giulio Caravagna—developed BASCULE, a statistical framework that uses Bayesian inference to combine existing knowledge and update the analysis as new data become available. The goal is to make the identification of mutational signatures more robust, enable the discovery of signals not yet catalogued and group samples into interpretable molecular subtypes. The study, published in Genome Biology, lists Elena Buscaroli and Azad Sadr as first authors.

In the DNA of tumour cells, mutations do not appear at random: they tend to cluster into recurring combinations. These patterns—mutational signatures—can be read as the cumulative effect of specific mutagenic processes. In other words, by looking at which types of mutations prevail and how they are distributed, it is possible to formulate plausible hypotheses about what has driven the tumour’s evolution.

In recent years, thanks to large genomic datasets, several catalogues of signatures have been proposed. However, catalogues built with different methods may not fully overlap, and the analysis can become difficult to compare or standardise.

BASCULE was created to address exactly this point: to use existing catalogues as an informative starting point, without giving up the possibility of identifying new signatures when the data suggest them.

The method adopts a Bayesian approach: rather than treating the analysis as a “blank page”, it introduces an initial body of knowledge (priors—i.e., plausible prior information) and updates it with the evidence observed in the data. This is especially useful when dealing with complex signals: it anchors interpretation to what is already known, while also making uncertainty clearer and helping recognise when something truly distinct emerges from previously catalogued signatures.

Once it estimates, for each sample, how much each mutational signature is “present” (in practice, how much it weighs in the observed mutation profile), BASCULE can also bring together samples that look alike, forming groups with shared characteristics. The idea is to turn technical information into a more immediate reading that helps recognise tumour subtypes and, when the data allow, connect them to clinical differences.

In the study—which stems from the output of an AIRC-funded project—the authors show that this approach can recover already known subtypes across different cancers and, in some cohorts where clinical information is available, identify groups associated with different outcomes. From this perspective, mutational signatures are not only a “description” of mutations: they become a tool to better interpret the biological history of the tumour and distinguish patient profiles.

“BASCULE,” explains Giulio Caravagna, Professor of Computer Science at the Department of Mathematics, Informatics and Geosciences, “is a tool that allows us to analyse a large number of patients at the same time, identifying new groups of tumours that share similar mutational signatures. This kind of approach underpins so‑called patient stratification in oncology—one of the most important steps for modern precision medicine. By working at the level of mutational signatures, we can catalogue our patients and pinpoint those subgroups in which DNA damage follows well‑defined rules.”

The study is the result of collaborative work that also involved Human Technopole (Computational Biology Research Centre, Milan), Area Science Park (Research and Technology Institute, Trieste) and the University of Milan-Bicocca.