This online industry talk will be given by Richard Cooper, Oxford Drug Design Limited.

Registration is free but required - click here to get the Zoom link!

Title: Guide-rails vs guard-rails: useful generative AI models in drug discovery

Abstract: The field of drug discovery saw widespread, early adoption of statistical models of drug candidate properties, in order to attempt to reduce costs and to increase efficiency of finding new medicines. In this presentation we focus on methods for finding and optimizing small molecules which target specific proteins.

There is sometimes a domain knowledge gap between statistical modellers and medicinal chemists which can lead to misuse of datasets on one side and abuse of mathematical methods on the other. We highlight a few of the challenges and often repeated mistakes in these areas, which need to be addressed to maximize the impact of ML/AI models. In particular, the value of unconstrained generative AI models that require extensive filtering of output (guard-rails) will be contrasted with models with designed-in constraints (guide-rails) that efficiently search a physically meaningful and accessible chemical space.