One main priority is to predict everything, i.e. all assays and endpoints where a reasonable number of observations are
2nd Annual AI in Drug Development Congress
Friedrich Rippmann Director, Merck
SPEAKER SPOTLIGHTS:
The Future of AI in Drug Discovery
Chris Morris Scientific Technology Council
Jonny Wray
AI is set to revolutionise drug discovery, with machine learning approaches able to process data efficiently and select likely-successful targets. With so many potential applications, there is significant discussion about how best to integrate AI into the drug design toolbox and how it could be utilised in the future.
Head, e-Therapeutics
Willem van Hoorn Head, Exscientia
Read on to learn what has most excited senior experts from Merck, e-Therapeutics, Exscientia and the Science & Technology Facilities Council in this burgeoning field.
Q1: What is your main priority in AI-based drug discovery in 2018?
Q2: What is your main priority in AI-based drug discovery in 2018?
Friedrich Rippmann, Director, Merck
Jonny Wray, Head, e-Therapeutics
Friedrich Rippmann, Director, Merck
One main priority is to predict everything, i.e. all assays and endpoints where a reasonable number of observations are available. Then make predictions easily available to all that are interested (and even to those that are not interested, so that they are at least aware of what has been predicted – before they embark on e.g. synthesis). Another priority area is to achieve objectiveness for more complex decision situations, like the decision on which molecule or series to advance further, and which to drop. Then piecing elements together to coherent workflows will be the next challenge. All this will not replace e.g. the medicinal chemist but make him very efficient.
Our main use of AI-based techniques is in data augmentation – expanding empirical data sets using AI based approaches. These data sets we then use in non-AI based computational approaches that are based on mechanistic modelling of biological processes rather than the statistical approaches common to machine learning. Our main priorities are to continue to improve the performance of our existing approaches, look to integrate and ensemble other novel approaches, and apply these concepts of data augmentation via AI to data areas not yet addressed.
In 2 years time we will have proven that many individual AI tools and AI-supported processes actually work, and in 5 years time the sceptics will be the enthusiastic supporters – having gained time for more creative work, and deeper reflection.
Chris Morris, Scientific Technology Council My own priority is the application of machine learning methods to cheminformatics. There is lots of good work applying ML to target discovery, but rather less in QSPR, and the challenges are somewhat different. Chemical data are expensive and often not shared, so datasets are smaller. This influences the choice of method. It has also created a temptation to report models that are actually overfitted.
Willem van Hoorn, Head, Exscientia AI in drug discovery is new and still rapidly evolving, but to call it unproven technology is simply not true. It may not yet be the finished product, but it’s here and it’s here to stay. I would like to see this dismissive attitude change and have more productive discussions about how to get the most out of AI, for instance is the current drug discovery process still optimal? Efficient flow of data is more crucial than ever, tools and processes to access data should be designed not only with human users in mind.
Chris Morris, Scientific Technology Council I see three big opportunities for progress: the transfer to industry of machine learning skills, the creation of larger open chemical datasets, and a deepening understanding of how and when deep learning succeeds.
Jonny Wray, Head, e-Therapeutics I think the key development in the field over the next 5 years will be greater clarification of where AI, and computational techniques in general, can add value across the drug discovery and development process. At the moment AI is being hyped and, in a number of quarters, a large amount of overpromising seems to be happening. Heartening, though, is that practitioners are very aware of the requirements, and limitations, when applying AI to drug discovery. I think the growing consensus
is that the augmentation of human expertise with AI based predictions will be the way forward. The greater appreciation of these aspects, dampening of the hype, and the focus on specific problems where AI driven approaches can help, will be where the field goes over the coming years.
Willem van Hoorn, Head, Exscientia There currently is a degree of hype that should have abated by then and AI in drug discovery will have become the new normal. Some AI-based drug discovery companies that now operate like a CRO will have become companies that discover their own drugs. So far AI efforts have been focused on compound design, target discovery, drug re-purposing, and more recently synthesis planning. All of this could be done before without AI, I am hoping a brilliant mind will have come up with something truly disruptive that requires AI and could not have been done before. These experts will be presenting at the 2nd Annual Artificial Intelligence in Drug Development Congress this September in London as part of our PharmaTec Series. For more information on the event, please contact Angela at:
[email protected]