The Top Barrier To AI In Drug Discovery May Surprise You

Source –

By Simon Smith

He nailed the presentation. It was March 2016. Naheed Kurji was pitching his AI drug discovery startup, Cyclica. The audience was a large group of scientists. It included senior scientists and heads of computational drug discovery and chemistry. These were the right people. Kurji hit the key points and stayed on message. The technology worked. Things looked good.

But then came the Q&A session.

“Where are your peer-reviewed publications?” asked a senior scientist. “Where is your prospective validation?”

Such questions seem obvious now. But Kurji, Cyclica’s president and CEO, wasn’t prepared. His team focused on building the technology and ensuring it worked. They didn’t consider what scientists might need to buy in. “It was a humbling experience,” he says. And it kept happening.

So they changed their approach. They focused on answering questions, providing thought leadership and educating the marketplace. “They were curious about artificial intelligence, what it is and what it isn’t, how it can be best applied, and its benefits and pitfalls,” he says. “They wanted a better idea of how our platform worked, examples through case studies, validation, both retrospectively and prospectively, and peer-reviewed publications.” Or else nobody took them seriously.

The results of two new surveys reinforce Kurji’s experience. In a survey of 374 life scientists, nonprofit innovation advocate The Pistoia Alliance found technical expertise as the most cited barrier. In a survey of 330 scientists that BenchSci conducted with the Science Advisory Board, we similarly found lack of knowledge the biggest barrier to adoption. Seventy-one percent of respondents said “education about the technology” would increase adoption.

Startups and pharma companies, take note.

Nearly Half Of All Scientists Are Unfamiliar With AI For Drug Discovery

We found that 41% of scientists working in drug discovery are unfamiliar with AI. This includes 15% who are very unfamiliar. Pistoia’s survey found that 8% know “next to nothing.”

Why aren’t more scientists familiar with AI? I have long worked in health tech. So I assumed a conservative culture or concern about data privacy inhibited experimentation. Not so: 18% and 16% of respondents said so, respectively. The top barrier in our survey was “lack of knowledge and expertise about the technology” (62%). Another was “lack of knowledge and expertise about available companies and tools” (42%). Pistoia also found technical expertise the most cited barrier, at 30%.

Without deeper knowledge, many scientists will be skeptical. “In drug discovery, 99% of new molecules fail to deliver new medicines,” says Nick Camp, an independent drug discovery consultant who spent 20 years working in pharma R&D. “So it’s really easy to be skeptical, particularly about new and unproven, untested, unknown technologies.”

The industry has also seen overhyped technology in the past. For example, says Camp, computational chemistry in the 1980s and combinatorial chemistry in the 90s. Both failed to deliver on their promise to deliver drug candidates more efficiently.

Education And Case Studies Are Key To Adoption

The solution is to impart knowledge and provide proof.

These findings build on a study by Edelman and LinkedIn on thought leadership. It found that 82% of business decision makers say thought leadership content increases trust in a vendor. Forty-five percent said it led them to award a company business.

This is particularly important for AI startups in life sciences. We contend with two unique challenges: Machine learning is at peak hype and our customers are professional skeptics. So it’s essential to do these three things:

1. Educate about technology, costs, companies and tools. These are the most common knowledge gaps. Explain how the technology works. Detail its cost, benefits and cost-effectiveness. Describe available companies and tools. And explain where your company fits and differentiates itself.

2. Confront disbelief with concrete, credible evidence. Because machine learning is at peak hype, expectations outstrip reality. The market is full of vaporware. Use this to differentiate. Be transparent about what’s possible with your solution today and what may be possible tomorrow. And use demonstrations and peer-reviewed publications to remove doubt.

3. Deliver relevant case studies. Large pharma and biotech companies rarely have an appetite for case studies with startups — especially when their technology offers a competitive advantage. The resulting lack of case studies is a barrier to adoption. Scientists want to see relevant examples. To help, consider creating case studies with academic labs and smaller biotech companies.

Kurji now prioritizes such tactics with Cyclica. Their thought leadership content includes case studies, validation notes, publications, webinars and blog posts. “Those same companies that once turned us down have now engaged us,” he says. “And many more pharma companies have actively adopted our platform.”

While such initiatives are key for startups, pharma and biotech companies also play a role, says Camp. “At the moment, there are probably some very interesting things going on in companies, but it’s hard to know what’s being done,” he says. “It’s ironic in a way. There’s much more open science going on now in pharma. It’s almost like we need that type of scenario for AI.”

It’s possible we’ll see it in 2018. Most companies plan to expand the use of AI for drug discovery this year. Fifty-nine percent of scientists told us their organization will do so, and 94% told Pistoia they’ll increase use of machine learning within two years.

In which case, many more startups may find themselves pitching to tough crowds. Let’s hope they’re now better armed for the Q&A.