For the very first time, a new drug created solely by artificial intelligence (AI) is going into human clinical trials. Exscientia, an 60-person strong AI startup based in Oxford, UK and Sumitomo Dainippon Pharma, a Japanese pharmaceutical company headquartered in Osaka are the forces behind this novel medication targeted at patients with obsessive-compulsive disorder (OCD). Traditionally, a new drug compound requires 4.5 years to arrive at the human clinical trial phase while AI only took 12 months. This will not only accelerate the speed of drug discovery but also tremendously reduced its associated cost.
According to the Financial Times, which reported the breakthrough last Thursday (30 January), Exscientia’s AI platform has a suite of algorithms that conjured and screened through tens of millions of latent molecules before deciding on the ones to be synthesized and eventually tested. The best chemical structure (i.e., DSP-1181) was deemed the best candidate in this case as it targets specific neuroreceptor responsible for OCD.
Andrew Hopkins, Molecular Biophysicist and Chief Executive of Exscientia described the company’s AI as having “record-breaking productivity” because it has a faster learning speed as compared to other conventional approaches. That’s why only 350, which is one-fifth of the usual number of drug candidates, were created and tested in the process. Hopkins believes the algorithms could also be applied to other drug targets against a wide range of oncology and cardiovascular diseases.
It’s not just about speed and cost
Indeed, speed and cost are probably the two major reasons to explain the pharmaceutical industry’s eagerness towards AI. Last September, the deep learning model developed by an American biotechnology company, Insilico Medicine and researchers from the University of Toronto took only 46 days to identify a potential new drug. A process which can take up to 10 years and costs as much as $2.6 billion originally. Nevertheless, researchers cautioned although AI looks promising at the moment, it’s still a long way before AI-designed drugs were deemed safe to be used on patients.
As such, some companies are keen on combining genomic and clinical data to generate smarter algorithms that aid in the generation of targeted drugs for oncology patients. At the same time, researchers will be able to know whether a treatment is working in the shortest time possible through blood draws, so that unnecessary side-effects can be avoided or reduced. In the long run, a similar approach can also be used to predict or even prevent a possible relapse.
Others focused on screening a large amount of existing medications and redesigned them into news one. For example, California-based AI company, Atomwise launched in the year 2015, a virtual search for available drugs that can be redeveloped to treat Ebola virus. Within a day, the company’s AI system was able to predict two drug candidates that can significantly reduce the infectivity of Ebola. A process which generally would take months or even years.
Furthermore, as mentioned in our earlier blog article, quantum computing may advance the building of virtual human beings and complete simulations of all physiological and chemical processes. In silico medicine or the use of technology to model, simulate or visualize biological and medical processes could render as an alternative to human clinical trials in testing new drugs.
A door for the outsiders?
In December 2018, Google DeepMind outperformed a conference of biologists gathered in Mexico. Its AI system – AlphaFold, managed to predict the structures of proteins based on their genetic codes with significantly higher accuracy than fellow human experts. As proteins’ shapes underlined their functions in our bodies, AlphaFold believed this would help scientists in uncovering new protein-based drugs. Its preliminary success raised a question among the audience at the end of the conference, can a group of “outsider” (i.e., Google DeepMind) comes into the pharmaceutical industry so quickly and readily because of AI?
Some experts compared this with predictive analytics, they believe the role of AI, regardless of how it’s being applied, is to augment the capabilities of scientists, not replacing or displacing them. An AI algorithm could go sift out drug candidates that researchers know are likely to have adverse side effects or unable to discriminate positive and negative biological effects as diseases progress. As such, human experts need to present in the refinement process, to ensure AI will generate the best results.
Traditional drug discovery process worked on a specific set of hypotheses, experiments, and evidences at one time, this creates a leeway for possible biases. AI methodologies, on the other hand, assists researchers to avoid unconscious biases by looking at data as a whole. At the end of day, drug discovery is a team effort, whether AI becomes smarter than us or not.