only a subset of genetically defined tumours respond to target inhibition. This likely reflects the polygenic and context-dependent nature of synthetic lethality, with effects often influenced by modifier genes, environmental conditions, and cellular states. These complexities highlight the need for artificial intelligence (AI)-based models such as the hypergraph neural networks that explicitly model higher-order genetic and contextual interactions.
This project builds on recent computational work showing that transcriptomic buffering characterises robust, highly penetrant synthetic lethal (SL) interactions, and that trigenic SL interactions involving one biomarker and two paralogs can be identified using machine learning models. The overarching goal is to define the biological principles governing polygenic SL interactions beyond pairwise and trigenic relationships, and how to use these principles to predict new polygenic SLs.
Project aims:
- Identify and quantify fundamental biological characteristics of polygenic synthetic lethal interactions
- Model these characteristics using artificial intelligence algorithms with embedded explainable AI, in order to predict and interpret new polygenic synthetic lethal interactions in a pan-cancer setting
- Identify and interpret polygenic synthetic lethal interactions selectively associated with defects in CDH1, RB1, BRCA1/2, and over-expression of HORMAD1, in breast cancer
- Contribute to the design of validation studies of high-confidence synthetic lethal interactions, in preclinical models of breast cancer
Experimental work will be conducted in collaboration with the Loizou, Lord and Tutt laboratories (at the ICR), with the most promising candidates advanced to preclinical studies for therapy-resistant breast cancers.
The project is expected to yield high-impact discoveries in both methodology and target discovery, while providing interdisciplinary training.
Candidates must have, or be on track to receive, a First- or Upper Second- class Honours degree (or a Masters) in a quantitative subject and have experience in machine learning or AI, and must have a basic knowledge of human genetics.
References
LORD, C. J. & ASHWORTH, A. 2017. PARP inhibitors: Synthetic lethality in the clinic. Science, 355, 1152-1158.
SCHATOFF, E. M., GOSWAMI, S., ZAFRA, M. P., FORONDA, M., SHUSTERMAN, M., LEACH, B. I., KATTI, A., DIAZ, B. J. & DOW, L. E. 2019. Distinct Colorectal Cancer-Associated APC Mutations Dictate Response to Tankyrase Inhibition. Cancer Discov, 9, 1358-1371.
FENG, Y., YOU, H., ZHANG, Z., Ji, R., GAO, Y. 2019. Hypergraph Neural Networks. arXiv 1809.09401
HAIDER, S., BROUGH, R., MADERA, S., IACOVACCI, J., GULATI, A., WICKS, A., ALEXANDER, J., PETTITT, S. J., TUTT, A. N. J. & LORD, C. J. 2025. The transcriptomic architecture of common cancers reflects synthetic lethal interactions. Nat Genet, 57, 522-529.
IACOVACCI, J., BROUGH, R., MOUGHARI, F. A., ALEXANDER, J., KEMP, H., TUTT, A. N. J., NATRAJAN, R., LORD, C. J. & HAIDER, S. 2025. Proteogenomic discovery of RB1-defective phenocopy in cancer predicts disease outcome, response to treatment, and therapeutic targets. Sci Adv, 11, eadq9495.
