Current Edition

Upcoming Events


Seeking to adopt AI, drugmakers search for right mix of talent

Sanjiv Patel is trying to bridge two worlds.

The CEO of two-year-old biotech Relay Therapeutics, Patel and his team aim to map how proteins move in a hunt for new ways to lock therapeutic molecules onto problematic proteins.

It’s an approach that requires Relay to blend tried-and-true drug discovery techniques with advanced computing like machine learning and artificial intelligence.

Often billed as a transformative technology, AI has to date had a more limited impact in the biopharmaceutical industry. Yet drugmakers, from biotechs like Relay to large pharmas like Novartis, are eagerly exploring its potential to improve a notoriously messy drug discovery process.

Recruiting a team that can work across the experimental and computational worlds, however, is easier said than done. Much of the top AI talent is employed outside of biopharma in areas like traditional tech and the emerging financial tech space, and those who are experienced don’t often come with a deep understanding of chemistry or biology.

“It’s a challenge to recruit and retain scientists with AI capabilities,” said Jeremy Jenkins, head of data science for the chemical biology and therapeutics group at the Novartis Institutes for BioMedical Research (NIBR). “Most candidates know more about life science or more about computer science; hybrids are extremely valuable in this industry.”

Relay’s Patel agrees, “We are taking two different cultures and putting them together,” he said in a separate interview.

Learning biologists’ language

Relay has set out to understand how protein motion relates to function and thereby design better molecules to treat disease. As one step in that process, the biotech runs thousands of computer simulations which allow it to create a kind of molecular movie. Machine learning then enables Relay to process and learn from all of those thousands of simulated trajectories.

But translating protein motion into data a machine learning algorithm can understand isn’t an easy task.

“If you hire someone who has been working on Alexa at Amazon and tell them I need you to come up with a representation of a protein to use in machine learning, they are just not going to grasp the fundamental nature of the problem,” explained Pat Walters, VP of Computation at Relay.

The biggest hurdle facing biopharma adopters of AI is finding skilled computer scientists who have relevant expertise to drug discovery.

“If they’re experienced in machine learning, they most likely were not in genomics,” said Christoph Lengauer, CEO of Celsius Therapeutics, a biotech focused on single cell RNA sequencing. “They were in finance, engineering, advertising, mathematics. The challenge is, how can we help those people learn the language that biologists speak as quickly as possible.”

Luring AI talent from tech

Biopharma companies, even the Pfizers and Novartis’ of the world, are also starting from behind in their pursuit of the top AI and other computer science talent.

“The best machine learning people are not in pharma today,” said Regina Barzilay, a professor at the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory.

Technology firms and other industries like finance all have an inside track in what CB Insights, a research firm, dubs the “AI talent wars.” Relay’s Walters notes that even people with experience in both AI and a field like chemistry or biology may still be attracted to go work for companies like Google or Facebook.

In the broader field of “big data,” demand for employees is estimated to widely outstrip supply in 2018, according to PwC, leading to increased competition among employers.

With companies chasing experience, annual salaries can easily reach six figures. Searching for artificial intelligence jobs in Cambridge, Massachusetts on the hiring site Glassdoor, for examples, turns up more than 100 openings in biopharma companies, including top firms like AstraZeneca and Sanofi. Most carry estimates of pay packages that range from $100,000 to $200,000.

Such searches, albeit unscientific, give a hint of the level of competition, too. All told, Glassdoor lists nearly 2,500 artificial intelligence-related job openings in Cambridge and more than 4,000 in San Francisco. Only about 250 across both cities are tied to biopharma firms.

“There is a race for talent, not only within our own industry but across many industries in this space,” said Julie Schiffman, VP of business analytics at Pfizer. “The human talent is as important as your AI algorithms to have success here.”

Come work for us

Biopharmas are aware of the uphill battle they face in wooing AI researchers. They may, however, hold an advantage in selling their corporate mission.

While the pharma industry at large suffers from a bad reputation over drug prices, biopharma can offer the chance to work with top scientific researchers on diseases affecting millions of Americans.

“The only way we can break in is with the promise that now artificial intelligence in combination with genomics can have an impact on people’s lives and eradicating disease,” said Celsius’ Lengauer. “If I’m young and experienced in AI, I might leave my advertising job and go into what we are doing.”

Companies and researchers are also trying to improve awareness and communication between the computational and experimental fields.

At NIBR, Jenkins says the pharma launched a company-wide goal to increase digital awareness and encourages its data science community to provide training to bench scientists.

From the academic side, MIT’s Barzilay is co-developing a class for the spring 2019 semester that will combine machine learning with life sciences, and recently co-led a summit that brought together representatives from industry, regulators and MIT.

“Within a few years, we will see more and more people who are fluent in both languages,” Barzilay said.

Biology still trumps software

As biopharma turns toward AI, however, researchers and executives in the industry say a more pragmatic view of what role the technology can play is needed.

AI-focused startups may boast of cutting drug discovery timelines from years to months or even weeks, but biology has a way of complicating matters.

“The idea that AI on its own will create new medicines is probably the wrong way to think about it. AI can’t do drug discovery by itself,” explained Mark Murcko, a Relay founder and formerly the chief technology officer at Vertex. “You need people with experience hunting for drugs.”