William McCorkindale

Research

Machine learning tools allow us to analyze datasets and discover which features are statistically relevant for making accurate predictions. By applying these tools to chemical databases, we can predict molecular properties, the products of chemical reactions, and more.

The topic of my research is to employ these tools specifically to advance the discovery of new drugs. Drug discovery is an arduous, multi-stage process involving the simultaneous optimisation of many attributes such as binding affinity, metabolic stability, and the ease of synthesizability, all while under time & resource constraints.

With this in mind, my efforts are focused on two particular areas: Firstly, the development of physics-based ML methods to improve early-stage screening hit rates and hence shorten the time needed to arrive at a viable lead compound. Secondly, improving the reliability of chemical reaction prediction models via principled removal of dataset biases and incorporation of physical reasoning, in order to decrease the cost of molecular synthesis in the design-make-test cycle.

I am applying these methods within the open-science COVID Moonshot initiative to develop a patent-free COVID-19 antiviral. I passionately believe it is vital that the tools I develop are applicable to the real world as opposed to being academic curiosities.