About the Project
Antibodies are widely used in medicines and diagnostic tests because they can recognise specific biological targets with high precision. However, discovering and producing new antibodies is often slow, expensive and dependent on cells or cell-derived machinery. Researchers are therefore combining laboratory screening with computer modelling and machine learning to identify promising binding molecules more efficiently, but these steps usually remain separate and require substantial human…
intervention. This project will instead create an automated cycle in which machine learning proposes short antibody-like peptides, chemical synthesis produces them, experiments test their performance, and the resulting data improves the next round of designs. The aim is to discover robust, inexpensive alternatives to antibodies without relying on cells for either production or selection.
Dr Baker’s and Prof Sosso’s current peptide development strategies utilise machine learning (ML) and molecular dynamics (MD) simulations in parallel to improve antibody-mimetic candidate selection. They have focused on sequence-based machine learning models and molecular dynamics simulations to predict peptide binders directly from antibody sequences, aiming to accelerate discovery campaigns by leveraging underused, small-scale datasets. On the other hand, Dr Parkinson’s research has focused on self-driving labs that undertake experiments without the intervention of humans. This project will bring together these research strands to develop a fully automated closed-loop (design-build-test-learn), thus allowing for rapid peptide drug development without requiring the intervention of a human.
In short, what we aim to achieve is a discovery framework that makes and tests its own lead compounds, fits into a single room, runs 24/7 and requires no human input. The antibody-mimetics produced will be low-cost, robust short chain peptides easily deployed in therapeutics and lateral flow diagnostics.
