Category: Data Analysis and Informatics
Phenotypic approaches have many advantages over target-based strategies, however they also come with key challenges. Specifically, de-convolving the target binding profiles of phenotypic hits is incredibly challenging, and without such knowledge optimizing compounds for phenotypic efficacy must often rely on trial and error approaches. By combining three deep convolutional neural networks we have developed an AI platform that can: 1) Automatically analyse complex imaging data generated in high-content phenotypic screens; 2) Rationalize how compound structures relate to phenotypic effects by leveraging vast existing experimental target-based datasets; and 3) Optimize compounds against multiple objectives including phenotypic effects through reinforcement learning approaches. Together we believe these approaches can solve many of the challenges facing phenotypic approaches.
Sam Cooper– PhD Student, Imperial College London, Emsworth, England, United Kingdom
Imperial College London
Emsworth, England, United Kingdom
Sam is completing his PhD with Dr. Chris Bakal and Prof. Robert Glen at the Institute for Cancer Research and Imperial College London. During his PhD titled ‘relating chemotype to phenotype’ he has developed image and data analysis pipelines for high throughput siRNA screens, live single-cell imaging studies, and analysis of chemical properties. Currently he is focusing on applying deep learning strategies to phenotypic discovery and compound optimisation.