Drugs cost too much, for patients, for insurers and even for pharmaceutical companies that design and sell them. A new drug can cost more than a billion dollars to bring to market, and that cost has to be borne by someone. New drugs increasingly target more specialized health problems, and with it, sub-groups of patients suffering from particular complaints. This leads to better treatments using more targeted therapies but requires more drugs to cover all ailments.
Increasingly, artificial intelligence (AI) is being used to improve drug discovery and development, delivering better drugs that are faster and cheaper to design. Drug discovery, the act of finding a plausible candidate drug, and drug development, showing that it is safe and effective, are arguably two of the most complex ventures in modern science, often taking 10 years from concept to clinic.
New combinations, new uses
Drug discovery is hard because the number of potential chemical compounds is enormous, due to the many different ways of combining atoms into molecules. AI is ideally suited to help navigate this space, highlighting good candidate compounds to test further. AI methods are data hungry and need lots of good examples to learn from. Fortunately, decades of drug discovery by pharmaceutical companies has resulted in large digital datasets of chemical compounds and their properties. Combining these with the latest, most powerful AI methods is already producing candidate drugs and reducing discovery pipelines from years to months. For example, an algorithm from Exscientia, a leading UK-based drug discovery company, recently beat teams of research chemists in a competition to identify the best potential candidate drug for motor neuron disease.
As well as picking new compounds from scratch, AI can help redeploy existing drugs to treat other conditions. Benevolent.AI, another startup, has built a database of how drugs work and interact using AI to analyze genetic data and hundreds of millions of journal articles and patents. From this, AI methods that are similar to those used by Amazon and Netflix to recommend products and shows are used propose existing drugs for redeployment on other conditions.
Once a candidate drug has been identified, AI can utilize the vast and complex data that we can now gather on patients for drug development. For example, a full genetic analysis of one person results in more than 90 GB of information, about the same as four days of high-definition video. Combining that with digitalized health records, data from wearables (like smart watches) and new blood biomarkers gives us a rich view of disease and health. Hidden in that data are insights into who is most likely to respond well to a new drug, and who might suffer from side effects. AI is able to make sense of this data, to extract these subtle signals from the noise. Thus, we can effectively treat sub-types of diseases, and find the people most likely to benefit from an intervention. This area of “stratified” or “precision” medicine has attracted multi-billion-dollar government backing in the United States and the U.K.
Of course, as in any scientific discipline, we need to ensure that the evidence base for medical decisions is accurate and transparent to external validation. This is particularly important for AI tools that use automated computer programs to mine for patterns in large medical datasets.
Through these coming developments, as data sets become richer in information and AI algorithms more accurate, we can look forward to a future where AI has made drug discovery and clinical trials more streamlined, more cost effective and more likely to succeed, giving patients and clinicians better treatments within their armory to maintain and improve human health.
Paul, Steven M., et al. “How to improve R&D productivity: the pharmaceutical industry’s grand challenge.” Nature Reviews Drug Discovery 9.3 (2010): 203.
