Team:Paris Bettencourt/Software
Software
Directed evolution is a Nobel prize-winning method for peptide engineering. However, it is largely limited by the depth of mutant library and the fitness landscape. Here we developed an artificial intelligence (AI) software, AMPDesigner, to overcome the difficulties and accelerate the design-build-test cycle.
Our software includes two components: a designing algorithm and a machine-learning model that guides the designing. The machine-learning model is trained based on the experiment of randomly synthesized 319,586 peptides. The designing algorithm searches for better AMPs from a starting sequence, based on machine-learning model’s predictions. It is conceptually similar to directed evolution, except that it evolves the peptide in silico, rather than in the falcon tubes.
Using AMPDesigner, we generated 12,000 mutants from a few natural AMPs, with variations in the position and density of positively charged residues, and then experimentally screened for the top candidates in a high-throughput manner.
paris-bettencourt-2018@cri-paris.org 
