[Recorded] The BII Lectures - Artificial Intelligence in Drug Discovery: What is Realistic, What are Illusions?
01 Jul 2021
[Recorded] Dr. Andreas Bender Artificial Intelligence in Drug Discovery: What is Realistic, What are Illusions?
-> Link to slides: here
** Please note that the materials are not meant to be modified/reproduced/circulated without permission.
Title: The BII Lectures - Artificial Intelligence in Drug Discovery: What is Realistic, What are Illusions?
Speaker: Dr. Andreas Bender; Reader, Molecular Informatics, Department of Chemistry, University of Cambridge, UK
Artificial Intelligence (AI) has had a profound impact on areas such as image and speech recognition; however, comparable advances in drug discovery are still rare. This contribution will explore some of the reasons why this is the case, starting firstly by embedding the application of AI in drug discovery into its historical context. Next, we will explore which types of improvements – namely to speed, cost, or quality of decisions – are able to have the most profound impact on the success of drug discovery projects, and we will see that it is quality of decisions in clinical phases that matters most for project success. However, while computational algorithms are tremendously powerful today, much of the data we have available in drug discovery projects comes from the preclinical, proxy domain – which possesses only limited utility to make predictions for clinical stages. Hence, while numerically we are at least to an extent becoming better at modelling proxy endpoints, this leads to a ‘models of models’ situation, which only translates to a very limited extent to the safety and efficacy of compounds which will reach patients in a clinical setting. Even more fundamentally, there are intrinsic aspects of chemical and biological data which make labelling data much more difficult than in the image- or speech-domain, such as its conditionality (e.g. the dependence of effects on dose), which pose conceptual problems to being able to apply computational algorithms in the drug discovery domain. The contribution will conclude with aspects of model validation and perceptions of AI in society which pose additional problems for their unbiased evaluation and development, and point out developments which are needed for applications of AI in drug discovery to contribute to the clinical safety and efficacy of future medicines.
Dr. Andreas Bender is a Reader for Molecular Informatics with the Centre for Molecular Science Informatics at the Department of Chemistry of the University of Cambridge, as well as a Director for Digital Life Sciences at Nuvisan in Berlin. He received his PhD from the University of Cambridge and worked in the Lead Discovery Informatics group at Novartis in Cambridge/MA as well as at Leiden University in the Netherlands as well as AstraZeneca before his current post.