The Future of Drug Discovery: AI Impacting Upon Hit ID Strategies
Eureka Staff

The Future of Drug Discovery: AI Impacting Upon Hit ID Strategies

Is AI revolutionizing how laboratories screen for lead molecules?  Here’s a CRO scientist’s perspective. The fifth in our series.

Grant Wishart PhD, Director CADD & Structural Biology, leads the Charles River CADD & Structural Biology groups, which provide computational chemistry, structural biology, biophysics and protein production services across partner drug discovery programs. One of the hottest areas right now is AI. Dr. Wishart’s group, through a partnership with Atomwise is applying AI technology to predict binding to protein targets for vast numbers of compounds including synthesis-on-demand libraries. These experiments can be done within a matter of days. Not surprisingly, the power of AI is sowing both hype and scepticism. Dr. Wishart believes demonstrated success stories on live drug discovery programs are the way forward in alleviating doubt as well as managing expectations. Eureka connected with Dr. Wishart as part of its series on AI in Drug Discovery to learn more about how AI is impacting our contract research. Here are his responses.

Eureka: How is AI changing how laboratories like CRL and its client screen for lead molecules?

GW: At Charles River we carefully examine our partner’s targets of interest and make informed recommendations on the most suitable hit identification strategies which may include screening, knowledge based and conventional virtual screening approaches. Our recently announced strategic partnership with Atomwise, brings Atomwise’s unique structure based AI technologies based on convolutional neural networks for hit identification and optimisation to our partner projects. The power of this technology is an exciting step forward for Charles River into the AI arena as it opens up the opportunity to predict binding to protein targets for vast numbers of compounds including synthesis-on-demand libraries within a matter of days. Access to such large regions of chemical space becomes very important for those protein targets that are traditionally considered as challenging for hit finding and highly competitive “hot targets” where access to novel chemical space is highly desired.

Eureka: Is this only useful in the search for small-molecule drugs or are we also using it for therapeutic antibodies and other large molecule drugs?

GW: AI technologies can be explored in virtually any situation where there is data that can be modeled. Machine learning and AI based approaches have already proliferated across all drug discovery modalities with reports of application in the design of therapeutic antibodies, and proteins. At Charles River our exclusive partner for antibody discovery, Distributed Bio integrates computational immunology, bioengineering and robotics to generate computationally optimized antibody libraries.

Eureka: It sounds like a game-changer for Early Discovery teams, but there have to be challenges. What are some of the biggest challenges in using AI software?

GW: The most ubiquitous challenge across the AI field in the Early Discovery space is access to vast amounts of quality data to generate models that are truly predictive in a prospective manner with a wide domain of applicability. Such data can either be public sources such as ChEMBL or the huge proprietary data sources that reside within Pharma companies. At Charles River, we are standardizing our data storage and processing to enable optimal data use in model exploration and analysis. As a services company, this data is most often owned by our partners. Therefore within individual partner drug discovery projects this data is being leveraged to advance the project and at a more holistic level we are working with our partners to seek permission, where appropriate to explore global modelling efforts.

Eureka: With that said, how big a part is AI playing in how Charles River identifies novel compounds that can hit their targets? Are you finding success?

GW: We are in the initial stages of applying AI in hit identification and optimization at Charles River so it is a little early for definitive data on the levels of success for our partner programs. However it is anticipated that the application of AI technologies in our organization will grow significantly in the near future and will have a major positive impact upon our ability to find hits for challenging targets. This is expected to result in quicker timelines to transition projects into hit-to-lead and lead optimization phases.

What do you think the biggest challenges in using AI to discover drugs?

I am not sure that AI will discover drugs but it will contribute to what we do in drug discovery and support the scientific teams responsible for delivering clinical candidates. For some organizations the capital investment to access AI technology may be a hurdle and the decision on building internal capabilities or accessing through partnerships with dedicated technology companies and/or CROs becomes an important one.

What do you think will be the “Next Big Thing” in the application of AI in drug discovery?

There will always be a “Next Big Thing” in drug discovery which is often over hyped and there many historical examples which do not quite deliver, combinatorial chemistry perhaps being the most cited example. Perhaps the next steps forward will be on the implementation strategy and how AI technologies at different stages of the process can be seamlessly integrated within organizations. However the challenge still remains to demonstrate the current and emerging applications of AI prospectively impacting upon the drug discovery process. A level of skepticism towards AI in drug discovery exists, primarily due to the associated hype, and this skepticism will gradually be alleviated through demonstrated success stories on live drug discovery programs.

Thanks for tuning in. Our sixth Q&A in this series on AI in Drug Discovery will be with John O. Mitchell, a theoretical chemist at the University of St. Andrews, where he uses theoretical and machine learning techniques in pharmaceutical chemistry, condensed modeling and structural bioinformatics. Mitchell will be discussing ethical dilemmas, namely privacy and job attrition, that AI raises.You can follow our series here.