As a cautious swimmer slips slowly into the cool waters of a giant swimming pool so do biopharma executives: They tread into the murky abyss known as artificial intelligence (AI).
The refreshing promise of its potential is vast. It could drastically speed up drug discovery and development, identify more well-defined biomarkers and endpoints, lessen costs, and bring life-altering treatments quicker to patients. It could do whatever a team of the sharpest human brains could never do throughout many years of trying.
But the technology of AI – an umbrella term that includes machine learning, natural language processing, big data and digital health applications – is unproven when it comes to biopharma therapeutics. There is a true risk of data drowning, the up-front investment is significant, and companies struggle to find skilled data scientists with biological expertise.
Nevertheless, the message at many industry conferences within the last few years is the same: This is the future.
"The programming languages or the algorithms that have been developed have hit an influx point where they're actually able to start to do things that we couldn't do before," said Joel Haspel, vice president of digital health, life sciences, at Clarivate Analytics. "The computers are able to start to gather their own insights." Clarivate just launched Cortellis Digital Health Intelligence, a data solution that covers deals, independent health app reviews, digital health news and the latest discovery, development and commercialization trends.
Although biopharma has lagged behind the retail industry in adopting AI technologies, digital health software deals tracked by Cortellis indicate a substantial increase in AI partnerships. Licensings and research deals, and mergers and acquisitions, between pharma, biotech, AI companies, nonprofits, academia and government, have risen by 179% since 2014.
Two other ways of analyzing AI deals also arrive at the same conclusion.
A second, more narrowed, look at BioWorld MedTech's coverage of AI licensings, joint ventures and collaborations between companies indeed follow the same upward climb over recent years, albeit more dramatic, rising from only six deals in 2014 to 31 last year, a 417% increase. At 26 deals with five months to go, 2019 appears to be on track to become the highest year on record. (See Volume of biopharma AI deals, below.)
Third, an analysis of more than 100 AI companies cross-referenced with deals tracked by Cortellis and AI deals covered by BioWorld show a large jump beginning in 2015, going from 27 the year before to 56, a 107% increase. Those deals reached a high of 96 last year, a 256% increase from five years ago. They are so far at 55 for 2019.
"I think the trend in utilizing AI is just going to increase, particularly in the clinical and regulatory space," said Rick Finch, Clarivate's global head of consulting, life sciences. "I think it's becoming more and more accepted across the industry to use the technology to take over repeatable tasks and to increase the speed and quality of decision-making."
While financial terms of most AI deals are undisclosed, there seems to be a wide span of projected values among those in which the money is announced, ranging anywhere from a few million dollars to $1 billion or more. Most of the billion-dollar deals are acquisitions, but not always. Most recently, in April, Foster City, Calif.-based Gilead Sciences Inc. entered a $1.05 billion collaboration with South San Francisco-based Insitro to use machine learning to discover and develop nonalcoholic steatohepatitis therapies. On the lower end, a month earlier, Summit, N.J.-based Celgene Corp. partnered with Oxford, U.K.-based Exscientia Ltd. in a $25 million deal for its Centaur Chemist AI platform to discover small molecules for oncology and autoimmunity. Regardless of the wide range, the overall values appear to be increasing, most strikingly reaching a height in 2018 with Basel, Switzerland-based Roche AG's acquisition of Flatiron Health for $1.9 billion in April and of Foundation Medicine Inc. for $2.4 billion in July.
From retail to drug development
The field of AI was founded by mathematician Alan Turing, who was featured in the 2014 film "The Imitation Game" for his efforts in breaking the German Enigma Machine's code during World War II. In the decades since, social media companies, like Google and Facebook, began driving rapid growth, creating "massive amounts of digitized data," Haspel said.
The data has led to chat bots, which are accessed by virtual assistants such as Amazon's Alexa, to programs that can recreate Star Wars' scenes with the color-patterning of Van Gogh's "Starry Night." At Clarivate, machine learning is used to classify clinical trials within Cortellis. Biopharma companies are using it for research and development and to streamline operations.
"The strategic importance of AI has become obvious to most organizations within the biopharma industry, and a majority of them will be looking for ways to rapidly progress in this direction," according to Dmitry Kaminskiy, founder and general partner of Deep Knowledge Analytics, an arm of Deep Knowledge Ventures, who spoke in May at Digi-Tech Pharma 2019 in London.
While a number of pharmaceutical companies have already entered the space through acquisitions, "a large percentage of the big pharma companies recognize that they need to do this" and it is snowballing, Haspel said.
Paris-based Sanofi SA and Basel, Switzerland-based Novartis AG have recently announced efforts to use AI in the supply chain to create efficiencies and lower risks. Earlier this year, Roche signed a $67 million agreement with Exscientia to discover preclinical drug candidates.
New Brunswick, N.J.-based Johnson and Johnson's Janssen Pharmaceutica unit signed on with Paris-based AI company Iktos in April of this year to accelerate small molecule drug discovery. Janssen Research & Development LLC formed a multi-year alliance with Cambridge, Mass.-based Nference in June to use AI to find novel targets and disease subsets, as well as to stratify patients and identify optimal clinical trial sites. Separately, Janssen was already using AI to find better ways to develop drugs to address Alzheimer's disease at the earliest stages.
Finding disease early with AI
"We need to open up our thinking a little bit," said Vaibhav Narayan, the company's vice president and head of data science and digital health solutions for neurosciences, who spoke at the Biotechnology Innovation Organization's annual convention (BIO 2019) in Philadelphia in June. "Amyloid and tau have dominated our thinking. A lot of our AI application and data machine learning has essentially centered around what would it take to do a disease interception trial in Alzheimer's, what would it take for us to intervene early enough that we can actually bend the curve."
The Michael J. Fox Foundation for Parkinson's Research is using AI to derisk the disease and to accelerate new treatments by generating cohorts of natural histories, following patients longitudinally, said Mark Frasier, senior vice president of research programs.
"We have evidence to show we can enroll and find individuals that are at risk for developing Parkinson's based on certain genetic or clinical features," he said at BIO 2019. "We are starting to collect data on those individuals and understand who might convert and develop symptoms of Parkinson's disease and who might not."
The foundation has worked directly with machine learning company GNS Healthcare, based in Cambridge, Mass. Iya Khalil, chief commercial officer and founder at GNS, said she hopes her company's application of mathematics to build system models, using supercomputer power, leads to the ability to map surrogate phenotypes and biomarkers to preclinical experimental systems in order to speed drug discovery.
"It took a foundation to kind of recognize that there was this big gap on the data side and actually make the investment up front without knowing what the economic downstream consequences are going to be," she said. Through the Michael J. Fox partnership, data scientists "found out we could come up with and learn synergistic combinations and markers from whole genome datasets combined with phenotypic data, and those could identify patients who are progressing quickly through the disease vs. those who may be on a slower trajectory."
AI adoption vs. company size and cost
The Tufts Center for the Study of Drug Development and the Drug Information Association, in collaboration with eight pharmaceutical and biotech companies, conducted a survey during the fourth quarter of 2018, which found that 57% of small companies, 74% of mid-sized companies and 88% of large companies are using AI within their organizations. Most were using it for clinical patient selection and recruiting, as well as AI-enhanced literature review, while a smaller percentage were using it for genetic data analysis, target identification, and to generate small molecule leads. Of those not using AI, they cited a lack of skilled staff, regulatory concerns, budget constraints and a mistrust of unvalidated technologies.
Although some small- and mid-sized biopharma companies are hanging back from investing in AI technologies, problems of rising costs, complicated diseases, stricter regulations and the pricing debacle, could make partnering a necessity. According to Kaminskiy, the cost to bring an asset to market has climbed from $1.19 billion in 2010 to $2.17 billion in 2018, yet forecast peak sales per asset declined from $816 million to $407 million during that timeframe.
Incorporating AI into a business model "requires a lot of work. There's a lot of effort to build the data, run the analysis. Who's going to pay for all of that stuff?" Finch said. "I think the conversation I hear is how do we take actionable insights from all of that data that's out there and share that across the ecosystem."
Cost concerns should be contrasted with the price to conduct a phase III trial, Khalil said at BIO 2019. "It actually is still a small percentage of the cost of trials today, which in [central nervous system diseases] where we end up with a lot of failures, it should be something that if it's designed from the beginning the system can afford to do, and hopefully increases our level of success, so the investment will be worth it."
Partnerships show fruit
Meanwhile, a number of AI companies have already signed deals with large pharma companies, with some beginning to show fruit. In its deal with Paris-based Servier, three-year-old Iktos was able to apply AI to identify a single molecule from 800 possible that met all 11 pre-defined objectives. Exscientia recently delivered to Glaxosmithkline plc in April a lead molecule for chronic obstructive pulmonary disease, using its Centaur Chemist platform.
The successes are leading to more partnerships across the landscape. In addition to Celgene's March partnership with Exscientia and Gilead's April deal with Insitro, Iktos signed on with Darmstadt, Germany-based Merck KGaA to use AI for three drug discovery projects. Merck already was working with Toronto-based Cyclica Inc. to use its proteome screening platform, Ligand Express.
About 56% of respondents to the Tufts/DIA survey said their AI partnerships were focused on drug discovery/preclinical and 42% were focused on supply chain. Only 6% said AI had contributed to the regulatory approval of a drug at their organization. The hope, however, is that the technology will lead to better development decisions, safer and more cost-effective drugs, and streamlined business operations.
"At the end of the day, I hate to say it, but health care is still a process of elimination," Haspel said. "These algorithms should help us drive more personalized medicine, more targeted drugs, and that then really changes outcomes."
Deep Knowledge Analytics estimates that there are about 350 investors, 150 AI companies and 50 pharma and tech companies focused on using the technology for drug discovery, biomarker development and advanced research and development. Of the 150 AI companies, 55% are based in the U.S. and just under 20% are in the U.K. Canada and the European Union each host about 9%, and Asia has 7%. Of the investors, 63% are U.S.-based.