Discovery Biology
- Build consensus disease signatures across public omics datasets
- Cluster and annotate cell types in single-cell atlases
- Map spatial transcriptomics onto tissue architecture
Group · Molecular
High-throughput biological data, read for mechanism — and carried forward. Our work spans the computational drug discovery pipeline: finding the disease signal in public omics data, testing with human genetics whether a target is causal, designing the molecule or biologic that engages it, and producing the safety and clinical evidence that decides whether it should be dosed.
The areas below are ordered the way a programme actually moves — a hypothesis becomes a target, a target becomes a molecule, a molecule becomes a dose, and a dose becomes evidence in patients. Projects usually sit in one area and borrow from the ones on either side.
Genomics, transcriptomics, epigenomics, proteomics and metabolomics each describe one layer of a tumour. Integrated, they describe a system. We synthesise multi-dimensional data from international consortia — TCGA, GEO, ICGC — to characterise tumour heterogeneity and identify clinically actionable targets.
Bulk RNA-seq gives a high-resolution view of the transcriptome across a sample. We use it for differential expression, isoform discovery, gene fusion detection and alternative splicing analysis across pathological states.
scRNA-seq resolves expression to the individual cell, which is where the interesting heterogeneity usually lives. Our work here concerns the tumour microenvironment, rare cell populations, and the cell-state trajectories that accompany progression and immune response. Spatial assays put those states back on the tissue, so a signature can be read against the architecture it came from.
Where a target emerges from the omics work, we ask whether human genetics supports it being causal — fine-mapping GWAS loci to genes, Mendelian randomisation for direction of effect, constraint and tissue-specific expression for safety. Targets that survive that gate are followed structurally, through binding-site characterisation, structure prediction, docking and stability assessment, and then into virtual screening, generative design against synthesisable routes, and in-silico binder and antibody design.
A candidate is only useful at a dose that works and does not harm. We model population PK and exposure-response to inform dose selection, predict off-target, hERG and secondary-pharmacology liabilities, and mine public pharmacovigilance data for adverse-event signals. Downstream, real-world EHR and claims data, survival analysis and systematic meta-analysis test whether the effect holds in patients.
Pan-cancer analysis reveals immunological and prognostic significance of CCT5 in human tumors, Scientific Reports, 2025.