Why artificial intelligence and pharmaceuticals are a perfect match

There’s something akin to Moore’s Law happening in data generation. Instead of transistors doubling in size every year though, data generation increases tenfold every two years. This is an idea corroborated by an often referred to piece of research by SINTEF that 90% of all the data in the world had been produced in the previous two years.

In the world of biomedicine, every day thousands of new scientific papers are published, not to mention new genome sequences, new biomarkers and new clinical trial data.

“To give you a sense of the scale of what happens every day, a life science paper is uploaded every 30 seconds; by the end of a whole day, you're talking about nearly 3000 pieces of scientific literature. One person couldn't read that in 10 years, let alone one day. And that's not counting all the other types of data: the genetic data, the biomedical imaging etcetera. We have to do things differently,” said Jackie Hunter, director at bioinformatics startup BenevolentAI, during a talk at Web Summit.

Despite the deluge of data available to scientists, they haven’t been able to convert this accumulated knowledge into new medicines, and new drugs for Alzheimer's, for osteoporosis, and for many cancers remain elusive. In fact, most people in the pharmaceutical industry never work on a drug that makes it to market because 90% of clinical candidates fail in development.

If any industry is an ideal candidate for artificial intelligence then it’s the pharmaceutical industry.

That’s certainly the opinion of Benevolent AI, who want to use AI to redefine how scientists gain access to and use the data available to them in order to stimulate innovation.

“What if a scientist could access all this data in a day or even less than a day. What if scientists could begin to pull all the data in from different sources, integrate it and really begin to answer new and important questions about disease,” said Hunter.

“Think of the impact on the industry, the impact on drug discovery, the impact for society and patients. This is something that we at Benevolent AI are committed to and we do every day.”

According to Benevolent, the current model used for drug discovery is broken, and industry leaders and governments who fund research are beginning to wake up to this high failure rate and lack of innovation. “The pace of innovation has not kept up with the pace of information, and that needs to change,” said Hunter.

What Benevolent proposes is taking all the information that can possibly be gathered – public information, proprietary information and information Benevolent pays to access – and using it to build a huge knowledge graph of relationships that are already known. With knowledge of these known relationships, the AI can then begin to identify relationships that may exist but have not yet been discovered.

“The analogy we use is that when you think about when the periodic table of the elements was built there were gaps,” said Hunter. “There were gaps because there were elements that should exist, but they were not yet known. What we're looking at is the analogy here for biomedical facts and data that should exist but are not yet uncovered, and we use that to discover new approaches to important biomedical problems.”

As well as providing better biological hypothesis in diseases, Hunter also claims Benevolent’s technology will contribute to making “better molecules much more rapidly”, which will make the process of drug discovery more efficient.

“When you have a hypothesis and you have a potential starting point in chemistry, you want to come up with a compound that will be a drug, and it needs to have certain properties: it mustn't be toxic, it must get into the right bit of the body, it must last for the right amount of time.

“What we can do by mining the corpus of the chemical information, is we can virtually screen many molecules, so in the end we only need to make and test about 10% of them, so we make 400 as opposed to 4000, and we get to a candidate in a year as opposed to two to three,” said Hunter. “This is a very exciting approach, I think it can really change the way we do drug discovery.”

Far from being a technology that may or may not influence the industry in years or decades, AI is a technology that is demonstrably affecting drug development right now.

Last year, one of the world’s largest pharmaceutical companies, GlaxoSmithKline, announced it would be investing $43m into the field, and other pharmaceutical giants including Merck & Co, Johnson & Johnson and Sanofi are also exploring the potential of using AI to help streamline the drug discovery process.

Benevolent itself sites an example of work it did on amyotrophic lateral sclerosis (ALS), which is also known as Lou Gehrig's disease or motor neurone disease, as an example of the efficacy of using AI in drug development.

ALS is a disease with no known cure that typically kills people within 2 to 5 years, although some patients, most notably Stephen Hawking, can live with the disease for much longer. Working with no prior knowledge of the condition, Benevolent scientists were able to generate a number of new hypotheses within a few hours that were worthy of further investigation.

The top five hypotheses identified were then taken to specialists in the field, working at the Sheffield Institute for Translational Neuroscience, to see what they made of Benevolent’s work.

“They were amazed because of those five hypotheses or potential new approaches to treatment we took, they had come up with one of them independently, but it had taken two years for them to come up with that hypothesis,” said Hunter. “They hadn't published it, and so they were quite amazed that we actually came up with that hypothesis independently, in a couple of weeks.”

Benevolent tested its five hypotheses, and while one didn't work, three worked as well as the standard of care currently available, and, promisingly, one worked very well indeed. That hypothesis was taken forward into a preclinical model of the disease where Benevolent was able to show that it would be capable of delaying symptom onset. The approach is currently being optimised, and hopefully, Hunter says, “we'll come up with a better molecule and we can move much more rapidly to the clinic”.

Success stories like that are the reason that AI will soon be a tool intertwined with the pharmaceutical industry.

That’s unsurprising considering AI seems to be able to solve problems that have perpetually plagued the field, notably the time it takes for drug discovery and clinical trial success to be achieved, as well as reducing the expense of reaching market, which the Tufts Center for the Study of Drug Development’s regular assessment of drug development costs pegs at just under $2.6bn.

Unlike other industries though, AI isn’t predicted to take over, but rather to augment the skills of scientists already in the field. “For me, this liberates myself and my scientists to be able to ask questions that we haven't been able to ask before because we haven't had this integrated data set,” said Hunter. “So I think this isn’t about replacing scientists, this is about augmenting scientists, allowing them to operate in a completely new way and actually really explore exciting new hypothesis in diseases.”

Given that AI offers savings in time and finance it’s little wonder that the industry is keen to take advantage of this paradigm shift in drug discovery.

“There's been a lot of hype about AI, and some of it may not be justified, but I think for healthcare, in general, and for drug discovery, in particular, AI is going to have a tremendous impact,” said Hunter.

“Without AI we will not be able to progress our new medicines and get those new medicines to patients faster, we will not have a sustainable industry and I believe if we embrace this technology and really open ourselves out to the new ways of working that it brings we will ultimately be of benefit not just to the industry but for patients everywhere.”

You can read more about the latest AI news in the technology magazine Verdict AI, where this article was originally published.