Artificial intelligence hit a big milestone in January — the first drug designed entirely using artificial intelligence entered human clinical trials. The compound, created by Oxford-based biotech company Exscientia, is aimed at treating obsessive-compulsive disorder and reached this stage in less than a year — five times faster than it usually takes to get a drug to this stage.
It is the speed that made the rest of the pharmaceuticals industry sit up and take notice. It takes on average 10 to 12 years to bring a new drug to market, and failure rates are high. The industry is desperate for ways to speed up.
The cost of bringing a drug to market has gone from $1.2bn to $2bn in the last ten years, but sales have more than halved.
“AI could be a bigger revolution for the pharmatech industry than the development of DNA sequencing,” says Andrew Hopkins, founder and chief executive of Exscientia. He believes that artificial intelligence could cut the cost of bringing a drug to market by 30%.
30% sounds like a welcome break for an industry stuck in a downward spiral of mounting costs and falling sales. Pharmaceuticals industry statistics make for frankly depressing reading. The cost of bringing a drug to market has gone from $1.2bn to $2bn in the last ten years, but sales have more than halved.
In part, this is because most of the easy drugs have already been discovered. What is left is to look for treatments for more complex and rare conditions. Searching for these is laudable but financially disastrous. Ten years ago, the pharma industry was making a 10.1% return on investment, last year it is just 1.8%, according to the latest Deloitte report on the industry.
No wonder then, that so many drug companies are keen to try out artificial intelligence tools. Exscientia developed the obsessive-compulsive disorder drug with Sumitomo Dainippon Pharma, but is also working with a number of other big drugmakers. The company, which began as a Hopkins’ research project at the University of Dundee, also recently signed an up-to-€240m deal with Bayer to research new cardiovascular and cancer drugs.
It also has similar partnerships with Bristol-Myers Squibb’s Celgene unit, Sanofi, GlaxoSmithKline and Roche.
A growing number of startups
Exscientia is not the only startup developing these tools. Hong Kong-based Insilico Medicine raised $37m last year after demonstrating that it could find a promising drug candidate for fibrosis and take it to pre-clinical validation in just 50 days.
Recursion Pharma, a Salt Lake City-based startup that uses artificial intelligence to search for drugs for rare diseases, raised a $121m round last year. Others to watch include Boston-based Berg, Cambridge-based Healx and Iktos in Paris.
There are many variants in the way artificial intelligence is used. The UK’s BenevolentAI, for example, scans data on existing drugs to find new uses for them. Exscientia’s specialism is using a combination of deep learning and neural networks to tackle problems even when there is very little data available.
“A lot of deep learning works best with hundreds of thousands of data points, but some of the most exciting problems in medicine are ones where there is very little data — for example, when a new gene has just been associated with a particular disease,” says Hopkins.
“It’s not even about saving time and money, but about boosting success rates.”
Budapest-based Turbine.ai, founded in 2015, on the other hand, has built a cell model that can be used to simulate how cancer works, with the aim of using artificial intelligence to find effective cancer medicines. In November, the company raised €3m in institutional financing, led by Delin Ventures.
“It’s not even about saving time and money, but about boosting success rates,” Szabolcs Nagy, Turbine’s chief executive, tells Sifted.
“You have an industry standard of under 10%, the success rate that we’ve seen in clinical studies,” he says. “Which is astounding if you think about how long it takes for a drug, a compound, to get to that stage — years and years. Only after all of that time do you find out that the drug doesn’t really work.”
The simulated cell that Turbine’s team has created consists of over 2,200 proteins and their interactions, constituting a dynamic representation of molecular interactions inside and between human cells. Researchers can use it to migrate the initial trial and error stages of drug discovery out of the lab and onto a computer, vastly speeding up the process of finding potential treatments.
Using the technology, Turbine is now able to run millions of permutations and then submit the most promising for clinical trials. “This means that from 10m we’re able to uncover the five to 10 really promising ones that then need to be checked out in the lab,” says Nagy.
“Biological simulation has been an interesting space but most previous attempts have failed to represent the deep complexity of biology,” says Vishal Gulati, a venture capitalist focusing on the convergence of data, healthcare and biology, who is an angel investor in Turbine.
However, he says that there is increasing interest in this space and Turbine is one of the most advanced platforms. “More recently, newer companies such as Cellarity in Boston have entered this space, which further validates Turbine’s approach.”
Will AI-drugs really work?
The speed gains from artificial intelligence look impressive, but industry experts caution that there is still some way to go before this method is proven.
“None of these drugs has gone all the way through clinical trials yet, so there is no proof yet for how well this works,” says Karen Taylor, director at the Deloitte Centre for Health Solutions. “Nothing has gone all the way through the system.”
“The tricky part about drug discovery is that we have huge areas of ignorance about human biology.”
Though a drug might perform well against a cell simulation like Turbine’s, an actual human body may still throw some surprises.
“The tricky part about drug discovery is that we have huge areas of ignorance about human biology, in both health and disease,” says Derek Lowe, a longtime drug discovery researcher who writes a blog about drug discovery and the pharmaceutical industry. Lowe adds that in many instances when drug trials fail researchers still don’t know why, “so how is anyone going to incorporate that into a model of the cell?”.
Deloitte’s Taylor adds that much of artificial intelligence drug discovery so far has focused on traditional small molecule drugs which tend to be simpler and more stable. It remains to be seen if they can move into more complex drugs, such as biologics.
Will AI take over the pharma industry?
What is interesting, however, is that many of the artificial intelligence drug discovery startups have not gone down the typical route of simply being acquired by a big pharma company once they get to an advanced enough stage of product development.
“I am surprised that these AI companies are not looking to be acquired,” says Colin Terry, partner at Deloitte. “They are so confident in their approach and methodology that they want to partner with many companies.”
“We built the company to challenge the culture of the pharma sector.”
If anything, many of the artificial intelligence drug discovery startups are raising money to take on a bigger part of the drug discovery process. Turbine raised money last year in part to build out its own pipeline. “Our big takeaway was that in order to really get a drug out there that’s going to benefit patients, we need to sort of go at this ourselves and start identifying targets,” says Nagy. “Once we’ve done that, then it will be much easier to convince pharma companies to go along.”
“Ideally, in the next round of funding we want to raise enough to be able to progress a really promising drug and bring to clinical trials ourselves,” he adds.
Exscientia, similarly, raised a €23m Series B round last year in part to help build its own laboratories to test the drugs identified by its software. Hopkins says he’s not looking for an early exit, he’s on a mission to change the world — or at least the pharma sector.
“My primary driver is to change the industry. We built the company to prove we could do it, and to challenge the culture of the pharma sector. We should not be satisfied with the current rate of drug discovery.”