Research

Tumours evolve through the accumulation of driver and passenger alterations, clonal selection and spatio-temporal dynamics. We develop algorithms and probabilistic models to reconstruct tumour evolutionary histories from bulk, multi-region, longitudinal and single-cell sequencing data.

Our work includes methods for the inference of cancer progression models, clonal lineage tracing, and evolutionary metrics able to predict long-term outcomes in multiple cancer types. We are particularly interested in linking evolutionary trajectories with treatment exposure and resistance mechanisms.

A recent example is ProgEvo / IPSS-M-Evo, an evolution-informed prognostic framework for myelodysplastic syndromes published in NEJM Evidence, which translates directional mutational trajectories into a refinement of an established clinical risk score.

Topics

• Progression models from cross-sectional and longitudinal sequencing data
• Phylogenetic reconstruction and repeated evolutionary patterns
• Clonal dynamics in solid and haematological malignancies
• Evolutionary metrics as prognostic biomarkers
• ProgEvo / IPSS-M-Evo – evolution-informed prognostic modelling and IPSS-M extension for myelodysplastic syndromes

Selected associated tools

• ASCETIC – inference of evolutionary signatures from genomic data
• LACE / LACE 2.0 – longitudinal models of cancer evolution from single-cell data
• PMCE – expressive models of cancer evolution with prognostic power
• TRONCO – inference of cancer progression models from heterogeneous genomic data
• VERSO – robust phylogenies and intra-host diversity for viral evolution

Mutational signatures capture the imprint of endogenous and exogenous processes acting on the genome. We design regularised matrix factorisation and statistical frameworks to discover and quantify mutational signatures in large cohorts of adult and paediatric tumours, as well as in viral genomes.

We are interested in how mutational processes relate to exposure, DNA repair deficiencies, immune response and clinical outcome, leveraging cohorts with tens of thousands of samples and multi-omics profiling.

Topics

• De novo discovery and stable estimation of mutational signatures
• Signature assignment in large pan-cancer datasets
• Mutational processes in haematological and solid tumours
• Viral mutational signatures and host–virus interactions

Selected associated tools

• RESOLVE – robust estimation and analysis of mutational signatures in cancer
• SparseSignatures – LASSO-regularised NMF for mutational signature discovery
• VirMutSig – discovery and assignment of viral mutational signatures

Tumour heterogeneity emerges at multiple molecular layers, including mutations, copy number alterations, gene expression, methylation and epigenetic states. We develop similarity-learning and clustering frameworks to integrate multi-omics data and identify clinically meaningful tumour subtypes.

Our approaches are applied to large-scale cancer cohorts to characterise subgroups associated with distinct biological mechanisms, prognosis and treatment response. We also investigate the robustness and interpretability of such clusters across independent datasets.

Topics

• Multi-omics similarity learning and integrative clustering
• Subtype discovery in solid and haematological malignancies
• Association of molecular subtypes with outcomes and treatment
• Dimensionality reduction and visualisation for high-dimensional omics data

Selected associated tools

• ASTUTE – genotype–phenotype mapping linking somatic mutations to gene expression programs
• CIMLR – integrative clustering of multi-omics cancer data
• SIMLR – similarity learning for single-cell and bulk genomics

A central mission of the TBL is to translate computational insights into clinically relevant tools. We collaborate closely with clinicians in oncology and haematology to design studies, analyse genomic and clinical data, and derive biomarkers and predictive models that can inform patient management.

Our work covers biomarker discovery, outcome prediction, therapy response, and integration of genomic, multi-omics and routine clinical data. We also explore the use of machine learning and AI models in decision-support scenarios, always with a focus on reproducibility and interpretability.

Topics

• Biomarkers and risk models in breast, prostate and haematological cancers
• Integration of genomic and clinical features for outcome prediction
• Evaluation of AI methods in clinical decision-making
• Tools for visualising and exploring patient-level genomic data

Selected associated tools and resources

• OncoScore – assessment of the oncogenic potential of genes from the literature
• Custom pipelines and models developed within clinical collaborations