Designing signalling bias: benchmarking AI methods for de novo peptide agonist design at the galanin GAL2 receptor
Supervisors:
Prof Irina Tikhonova, Queens University Belfast
Prof Graeme Milligan, University of Glasgow
Prof Andrew Jamieson, University of Glasgow
Project Summary:
Artificial intelligence can now design new proteins from scratch, and the first designed peptides that activate cell-surface receptors have recently been reported. But designing a molecule that binds is not the same as designing one that does the right thing: receptors signal through several pathways, and the therapeutic value of a drug often depends on which pathway it switches on. Nobody has yet designed that property deliberately.
This project asks whether signalling bias can be designed. It focuses on the galanin GAL2 receptor, for which structures exist with two natural peptides of opposing behaviour - one activating both signalling routes, one activating only a single route - providing an unusually clear template for what biased binding looks like.
Based at Queen’s University Belfast, the student will first benchmark the leading AI design and structure-prediction tools against well-characterised peptide receptors to establish which methods can be trusted, then use the best of them to design selective GAL2 receptor peptides. At the University of Glasgow they will synthesise those peptides and test them experimentally, measuring binding, signalling and selectivity, and feeding the results back into the next round of design. The result is training across computational design, peptide chemistry and molecular pharmacology.