The $1.2 trillion pharma industry has ambitions for AI technology, but ‘people are ignoring’ a lot of its uses

Samantha Lee/Business Insider

Drug discovery is one of the most significant areas of promise for AI in data-rich healthcare, but experts warn that too much hype is risky.
Biotech Moderna has since its launch taken a tech-forward approach to health solutions, including developing personalized therapies, and modeling variables such as clinical trial duration.
This article includes an overview of AI in health, including the top three trends to watch, and how biotech Moderna applies AI to free up scientists to make discoveries.
Read how AI is transforming retail, transporation, consumer technology, and more in other articles from our special report, How AI is Changing Everything.

For the pharmaceutical industry, which is staked out on the premise of finding, developing, and ultimately selling advanced new medications, there is no greater challenge than the sheer unpredictability of that task.

Technology known as machine learning and artificial intelligence could change all that, its boosters say.

That’s because in an industry with massive data sets at its disposal, machine learning and AI tech is able to ingest and make sense of it all, informing better predictions about complex subjects like which experimental drugs are most likely to successfully treat patients. In the pharmaceutical industry, where productivity from the research and development activities that lead to new drugs has been on the decline for decades, that’s desperately needed.

Swiss drug giant Novartis and its CEO Vas Narasimhan have touted the potential of machine learning and AI — with executive Jay Bradner telling Business Insider that AI is "the next great tool" to find new medicines early this year — before pulling back some more recently. Sanofi, which recently named its first chief digital officer, is using AI to develop digital therapeutics that it hopes can help with conditions like depression. The biotech Moderna is using it to make personalized cancer vaccines custom-crafted for each patient.

But this is also an area where hype is high. Critics say that talk is at odds with the tech’s accomplishments. They note, for example, that no one drug has yet been developed using AI.

"I think it’s really important in this space to avoid excess hype. A lot of people fall into, ‘this is going to change everything and discover drugs in 12 months.’ Nonsense. It discredits the field when people do that," says Daphne Koller, who has a background as an AI researcher at Stanford and serves as founder and CEO of machine learning startup Insitro.

Insitro is developing better models for which to test out new drugs, and as part of that recently announced a partnership with biotech Gilead on drugs for a disease that can lead to liver failure.

But Koller, who previously served as chief computing officer of Calico, Google’s life-extension spinoff, also says it’s unrealistic at this stage to expect an AI drug, given that the drug development process can take 10 years.

In the biopharmaceutical industry, "we’re just starting on this journey," she says.

The perils of hype were seen with IBM, which famously declared that its AI system Watson could better treat cancer. But the computing company’s efforts to apply AI in healthcare have fallen short of expectations, according to reports from STAT and The Wall Street Journal.

IBM also stopped selling a Watson-based product for use in pharmaceutical drug discovery because of subpar financial performance, STAT reported in April.

Lauren F. Friedman / Business Insider

Top 3 opportunities for health and AI

Biopharma companies’ operational and planning needs: Pre-launch

The biotech Moderna’s senior director of informatics, Dave Johnson, makes the compelling case that smaller problems at drug companies, like internal planning for operations like manufacturing, are getting overlooked when it comes to this technology.

Discovering new drugs: Growing

Drug companies and new startups are using this tech to find out which chemicals have promise as medicines, because it gives them the firepower to screen one million or more compounds, and predict which aspects of them could be useful.

Securing data and making predictions from images: Mature

Images of things like scientific experiments have always been important sources of information for biopharmaceutical companies, but machine learning has given the industry new tools for collecting data from imaging and making predictions based on that about, say, how a chemical compound works on cells in the body, or whether a given molecule might be unsafe in the body.

Moderna uses AI to crack tough business decisions and aid development of personalized medicines

For Moderna, the Cambridge, Massachusetts-based biotech company that went public last year in the biggest biotech IPO in history, a tech-driven approach to medicine is at its very core, from the tablets that monitor operations on the floor of its Norwood, Massachusetts manufacturing facility, to the algorithms that are used in its experimental cancer vaccines.

AI is being used across the biopharmaceutical industry to take on some of its most persistent challenges. But Dave Johnson, Moderna’s senior director of informatics, says that "the success we’ve had has been not focusing on these large moonshot programs but all the little things that make it really hard for our scientists to do their jobs" so they "can focus on their core jobs of being innovative."

"I think there are a lot of applications of machine learning, AI, that people are ignoring," he said. Those smaller problems, which are achievable, get overlooked, "and I think a lot of companies in the industry are going to struggle with that."

Moderna has been able to see and seize those opportunities because the company digitized all its operations from the very beginning, which allowed it to generate data enabling the use of algorithms, he said.

For instance, Moderna is developing experimental products called personalized cancer vaccines that are being tested out for conditions like melanoma, and must be custom made for each patient. Because of that, and because patients enroll in research trials at variable times, planning the manufacturing process for the vaccines, from capacity for orders to even aspects like how much raw material the biotech needed, was especially complicated. The biotech also wanted to know what would happen if aspects were changed, and how much that would affect the product.

So Moderna turned to an algorithm, feeding in metrics from research trials to simulate hundreds of trials with thousands of pretend patients, asking questions like, "What if we change this timeline from five days to three days? What would that mean?" Johnson said.

The results allowed Moderna to match its manufacturing capacity to its needs, including through making large, crucial decisions about where to invest capital funds. "Before we had this tool, they were kind of guessing," he said.

The machine learning-based approach allows Johnson and team to supply decision-makers with concrete metrics, like what percentage of patients will need a particular dose. Those, in turn, go into designing a research trial and ensuring capacity is large enough to meet patient needs, without spending so much on capital that expensive manufacturing capacity gets wasted.

At Moderna, digital initiatives are emphasized at every level, from the company’s Chief Digital Officer, Marcello Damiani, to the amount of internal resources that are available, Johnson says.

This particular project began when Moderna’s clinical team, which takes charge of all matters related to its research programs, was having difficulty figuring out where to begin.

Johnson’s group analyzed both the problem and how achievable it was. It also considered the project’s value, or how great the return on investment could be. Quantifying that was hard. But because the project involved large capital decisions being made around the company’s future, "it was clear that it was an overwhelming return on investment," Johnson says.

So the team set it into motion, with work beginning in February of 2018, all done in house. The initial release came about a month later. The algorithm has been used consistently in the time since for various projects around the company, including to plan for Moderna’s phase 2 research.

Because this project addressed a more or less straightforward problem, and could be handled with a simple algorithm, trained using a forecasting technique and the right data, Johnson doesn’t recall any large hurdles.

But that hasn’t been the case with every algorithm. The duration of AI projects can be highly variable, depending on their complexity and the availability of relevant data, according to Johnson. One especially tenacious, complex problem, for instance, has held up one project for years.

The AI solution Johnson and his team devised has been working and being used to make key decisions across four different teams at the company, including the manufacturing unit, he says.

That can be tracked easily, because Moderna came up with a plan and can follow whether there are any discrepancies as research develops.

Johnson says that "if there was a mismatch we would absolutely see that, and so far we haven’t seen those challenges."

Cody Glenn / Contributor

"I think much of what machine learning can do is aligned with what has already been done, but making it faster, cheaper…and perhaps more importantly, increasing the probability of success."

-Daphne Koller, Founder and CEO, Insitro.

NOW WATCH: An animated map shows every total solar eclipse around the world until 2040