01 Jun 2022
"Machine intelligence is the last invention that humanity will ever need to make” – Nick Bostrom.
In ancient Greece, Hippocrates made early attempts to introduce scientific rigor to the “art” of medicine. Numerous intellectual contributions over the subsequent thousands of years have shaped medicine into the scientific field we know today.
The process of drug discovery has evolved into an immense field of investigation, characterized by highly complex, time consuming, expensive, multidisciplinary processes carried out by a myriad of local, domestic, and international organizations.
Drug discovery and design have come a long way since the days of iteratively applying known naturally occurring toxins (e.g., those found in fungi or plants) against disease targets until a therapeutic effect was observed. Biology is fast becoming digital, driven by rapid deflation in the costs of gene sequencing.
However, large amounts of machine-readable data present both an opportunity to glean new insights, and a formidable challenge, as deriving these becomes increasingly harder with ever-growing volumes of data. It is also becoming more difficult to maintain a broad overview of major developments in adjacent research fields, which can often be useful in creative drug design. Recent studies have highlighted this challenge, estimating that almost 80% of medical data remained unstructured and untapped after being created (Kong, 2019).
Having delivered significant progress in other markets like cloud computing and cyber security, can AI play a valuable role in drug discovery?
In theory, drug discovery and design are exactly the sort of problems that ought to lend themselves well to the input of intelligent automation. For example, the number of potential drug molecule permutations is around 1060, presenting an attractive optimization problem for AI, which can be trained to recognize potential lead compounds and provide validation of the drug target and drug structure design. Notably, this task can be both forward- and backward-looking.
The power of AI is illustrated by the fact that in just four days, a team of researchers from BenevolentAI identified baricitinib as a potential Covid-19 treatment. The Eli Lilly drug normally used to treat rheumatoid arthritis could tackle both the Covid-19 virus, and the body’s inflammatory reaction to it. It was the first time AI had discovered an existing drug to re-target a new problem.
Having observed its tangible benefits, many companies are taking advantage of intelligent automation. E.g., as recently as 2020, Pfizer did not have the means to automatically screen one of its libraries holding data on 4.5bn commercially available compounds. Now, it can scan the entire database in 48 hours, vastly speeding up its ability to identify potential new medicines.
According to Deep Knowledge Analytics (2H2019), there were over 170 AI-powered research and development (R&D) companies globally, and 35 major R&D centers utilizing AI. A 2019 Deloitte survey highlighted that over 40% of drug discovery start-ups use AI to screen chemical repositories for potential drug candidates, 28% use AI to find new drug targets, and 17% use it for computer-assisted molecular design. Miraz Rahman, a professor of medicinal chemistry at King’s College London, believes that within the next decade, all big pharma companies will have integrated AI into drug discovery.
Importantly, there is little expectation presently that AI replaces human expertise. Rather, AI is seen as a way of enhancing it. Subject matter experts are critical in defining data for AI analysis and providing peer review and end-to-end verification of results. Furthermore, just as with any powerful tool, if left unchecked, AI can be used for nefarious means. In a recent demonstration, an AI model was trained with a starting set of molecules and tasked to calculate how to adapt them to become increasingly toxic. The outcome was troubling: within hours, the model had proposed over 40,000 potential harmful molecules.
Drug discovery is just one phase in the broader process of getting new drugs approved. Just as with the discovery, other parts of the process bring idiosyncratic inefficiencies that could be improved. Gene sequencing continues to better in speed, accuracy, and cost. Illumina is the dominant player in the industry, but newcomers like Oxford Nanopore continue to drive innovation. Cell manipulation is seeing dramatic advances driven by companies like Berkeley Lights. Elsewhere, Genmab has delivered excellent progress in the field of antibody screening. Finally, CROs like Icon allow large pharma companies to outsource some of the heavy lifting and focus on the most complex research. Each part of the value chain is seeing improvement, contributing to a greater whole.
Drug discovery remains an important part of the overall drug development process, and one that potentially lends itself well for trained automation to play a greater role in. With the right balance between human and machine, and the appropriate checks in place to ensure strict adherence to the task, AI looks set to play an increasingly important role in how we go about discovering new medicines. The future of drug discovery looks bright – Hippocrates would be proud of how far we’ve come.