Artificial Intelligence Led Selection of Crystal Conformations

The main objective for this project is to apply deep neural network methods to the prediction of crystal conformations. In order to achieve this, new conformational descriptors will be developed and computed in all conformations in the Cambridge Crystalographic Database (the world’s largest crystal database) as well as other unobserved conformations generated computationally. A neural network model then will be trained and used for the prediction of crystal conformations. This neural network model will then be applied to AZ compounds. If successful we will then go on to investigate deep reinforcement learning methods to enhance the predictive power of current prediction algorithms.

This project is jointly funded by Astra-Zeneca, The University of Manchester, and The Alan Turing Institute.