Exscientia, located in Great Britain, is at the forefront of Artificial Intelligence (AI)-driven drug discovery and design. By fusing the power of AI with the discovery experience of seasoned drug hunters, Exscientia is the first company to automate drug design, surpassing conventional approaches. Exscientia’s platform enables breakthrough productivity gains as well as new approaches to improve drug efficacy. Novel compounds prioritized for synthesis by its AI systems simultaneously balance potency, selectivity and pharmacokinetic criteria in order to deliver successful experimental outcomes.
Exscientia’s CEO is Prof. Andrew Hopkins. The company was established as a spinout from Hopkins’ laboratory at the University of Dundee. Hopkins spent 10 years at Pfizer, where he was responsible for establishing new research foci including the concepts of druggability and network pharmacology. After leaving Pfizer he took the Chair of Medicinal Informatics at the University of Dundee where he launched a number of initiatives and raised over $50 million for academic and commercial research activities. He is the author of some of the most highly cited papers in modern drug discovery introducing concepts such as the druggable genome, network pharmacology and chemical beauty.
WuXi AppTec Communications, as part of a new industry series, recently interviewed Hopkins about the clinical direction and goals of Exscientia, as well as what the future looks like for AI in drug development.
WuXi: How would you describe the landscape of companies involved in this new field of AI? Are they mostly start-ups, or are the big data companies, such as Google and Intel, the dominate players?
Andrew Hopkins: From what we see, the most rapid developments are coming from groups who can fuse the technology aspects with domain expertise. As a consequence, it is not the size of the company that is important but the constituents of the team.
WuXi How would you describe AI, or machine learning, to the general public and to patients?
Andrew Hopkins: I would describe AI as an opportunity to solve well defined tasks more reliably and efficiently than currently possible. In the area of medicine, most people will only encounter the drug discovery world as a patient. As a patient, you want the best possible, timely diagnosis and as individuals we have experienced the time taken to move from an initial GP appointment to a follow-up meeting with a specialist. It is probably still some way in to the future, but it is reasonable to anticipate GPs being enabled by AI systems that help the GP to progress cases to the appropriate specialists.
WuXi: How does your company differ from others using AI in drug discovery and development? How are you applying AI?
Andrew Hopkins: Our company is applying AI to design new drug molecules. However, a key component of successful delivery is that we combine our AI technologies with the expertise of seasoned drug hunters in a ‘Centaur’ approach. Through this approach, new molecules are evolved and proposed for synthesis and analysis in a tightly aligned Design-Make-Test cycle where small batches of compounds are synthesised and tested so results can be rapidly fed back into the AI-driven systems. Our approach has already demonstrated the potential to improve the time required from project start to candidate to just one quarter of the industry average, delivering significant time, and cost and opportunity enhancements.
WuXi: What are the major benefits of using AI, or deep learning, in drug discovery and development?
Andrew Hopkins: The major benefits of AI are to systematize work that has to date been extremely manual and whose results have been highly dependent on the expertise of those put on the project. The application of AI means for example that it will be easier for an expert in one drug target family to move to a new target family without the usual decrease in productivity whilst the idiosyncrasies of the new project are learned.
WuXi: How will AI change drug discovery and development, and clinical research?
Andrew Hopkins: AI holds the potential to increase the speeds of project execution and thus potentially the capital efficiency of drug discovery by enabling better decision-making by bringing to bear all available information onto a design problem. The advantages, however, only appear when human and machine decision making are integrated into the right processes, rather than simply by adding new technology to an existing incumbent process.
WuXi : What are the limitations in applying deep learning to AI drug discovery and development?
Andrew Hopkins: The key limitation is that you require domain expertise to know what a reasonable question to ask is. There are cases where the questions being posed are too ambitious, the associated data insufficient and the AI question poorly constructed. In such cases, it is unfair to expect an AI system to identify a trend or put forward a proposal or to criticise when it doesn’t. We can already see from the huge efforts and investment required to deliver a reliable autonomous vehicle that the grand questions are not easy and they have enormous amounts of data to work on. With drug discovery, data are relatively sparse, which gives an extra dimension to the challenge.
WuXi: How will this field evolve over the next five years?
Andrew Hopkins: Without a doubt, more people will apply AI based methods to more applications. Perhaps not all will be successful, but from that set there will certainly be winners. We will get to a stage where experts are comfortable considering AI as a default approach to tackle well defined problems. As such, more effort will be spent defining the problem whilst letting the AI systems provide the proposals that the expert can then scrutinise. However, we are confident that AI-driven ‘centaur’ approaches will start to dominate. AI won’t replace drug hunters, but drug hunters who use AI will replace those who don’t.
WuXi: Will AI applications eventually become the norm for biotech and pharmaceutical R&D? If so, how soon?
Andrew Hopkins: AI applications will definitely become the norm for solving certain well defined problems within drug discovery. The technology is now at a tipping point where, when married with substantial data, conclusions can be drawn that are equivalent to researchers with many years of training. However, this does not replace those people; it makes them more productive as they can now spend more time on the more challenging strategic issues.
WuXi: What are the challenges/barriers in employing AI in pharmaceutical and biotech drug development?
Andrew Hopkins: AI can be employed in many different ways so there will be many valid ways answer this question. But with respect to our own technology, the challenge lies with pharmaceutical or biotech companies understanding that new technologies also demand changes in work practices to deliver their fullest effect. This is seen all over the world in manufacturing, warehousing and other areas, yet to date the pharmaceutical world appears to have been relatively immune to such change.
WuXi: What kinds of partnerships are important for the development of your company?
Andrew Hopkins: An ideal partnership is one where the collaborator looks beyond the AI technology and understands its full implication for drug discovery as a whole. If we are able to deliver candidate molecules sustainably in just one quarter of the time and with associated efficiency gains, then a partner should be able to see how this could have a substantial, positive impact on their clinical pipeline.
WuXi: How is your business model different from traditional biotech and pharma start-ups?
Andrew Hopkins: To date, we have developed Exscientia outside the usual industry norms for how a biotech is usually developed. We focused on doing deals with industry from day one, as it’s only by innovating ‘at the coal face’ by working on real drug discovery projects that the technology can truly be developed and validated. We have now begun to scale, with our first investment of €15 million from Evotec to accelerate growth. The question that is on our mind now is one of scale. We expect the story of Exscientia to deliver business model innovation as much as technology innovation.
WuXi: Will AI start-ups evolve to become drug developers themselves or will they focus on providing their technology in partnership with drug developers? As a follow-up, are existing pharma and biotech companies integrating AI into their R&D operations or will they rely on start-ups and big data companies.
Andrew Hopkins: Our company values working with collaborators, but we are also developing our own portfolio of compounds. Our systems are highly scalable so we do not want to be constrained by only working on targets that are selected by partners. A key challenge of introducing a new technology that changes a substantial existing process is that it may also require the introduction of different work practices. With our AI drug design technology, for example, we need a tightly integrated experimental component to deliver a tight design make test cycle. So, a key decision on whether to implement internally or rely on external companies will be the speed at which they can adjust to the new opportunity.