Big data combined with AI technologies has the potential to unlock knowledge and fuel innovation. That requires a data strategy based upon the right type and density of data. If the data set is too small or too thin, the strategy is likely to fail, according to panelists speaking at Biotech Showcase in San Francisco January 9.
The panel was moderated by Steve Dickman, CBT Advisors, and featured Ron Alfa, Recursion Pharmaceuticals; Gary Kurtzman, Safeguard; Katherine Merton, J&J Innovation; and Ajit Singh, Artiman Ventures.
“The bigness of ‘big data’ comes from dimensionality,” says Singh, managing director and general partner, Artiman Ventures. “Imagine one billion markers for one person, or 23 markers for a half million people. They’re both big data. One biomarker that applies to everyone on the planet, however, is not big data.”
The database speaks to the veracity of the data. If doctors don’t trust the data, they won’t use it. Therefore, “we need huge data sets at the population health scale,” insists Merton, head, JLABS NYC, Johnson & Johnson Innovation. When those are in use, “we’ll make progress.”
A lot of data already is available, but isn’t fully utilized. Commercial data, for example, is rarely shared with discovery groups. Yet, speaking as one who was once on the discovery side, Merton says, “Greater trust of commercial data would have saved us a lot of time.”
“When you have new knowledge, you will have a large challenge convincing the medical establishment—when the results can be tested,” Singh points out.
Oncology is probably the most likely field to accept new insights because of the seriousness of the disease, panelists agreed, but even it throws roadblocks to new approaches. Singh recounted a situation in which AI was used to evaluate tumor and normal tissues from 2,800 individuals to learn more about cancer. “The study indicated that healthy cells revealed more about the tumor than did the cancer cells,” he says, yet pathologists are trained to focus on the cancer cells. Consequently, they didn’t accept the data.
What makes a data strategy?
More and more companies see the wealth of data available and know they need a strategy to use it. Often they’re unsure what, exactly, that data strategy should look like.
Merton says, “The key is to develop an end-to-end strategy that clearly shows a progression. The pharma industry, however is (siloed) into an active commercial section and a relatively silent discovery portion,” when it comes to data. “To improve their drugs, they need to bridge that gap.”
Recursion Pharmaceuticals’ data strategy is based on three pillars, Ron Alfa, VP, discovery and product, says.
- It builds an extremely large data set populated with high-dimensional data, which lets researchers use computational tools to extract value.
- It generates data sets that are interrelatable from week to week. “Each experiment this week can relate to experiments we will run one year from now, even though the compounds and diseases are different. In our case, they’re interrelatable based on their stains,” Alfa says.
- It builds a foundation of data that has strong confidence levels and augments it with publicly-available data to add value.
Opening the black box
One of the greatest challenges to the routine use of AI is communicating what the algorithms actually do, Alfa says. “We can take an approach that admits we don’t know which features the algorithm identifies, or we can bring known controls to predict the actual biology (thus supporting the AI’s findings).” He favors the latter.
Singh also recommends giving physicians information and letting them make the decision, tempered with their own experience, regardless how the AI system derived the information.
Although there’s interest in the biotech industry in first using AI to duplicate what physicians already can do, Alfa says, the most interesting areas are in generating new knowledge for situations in which no one has tried to make diagnoses based on particular samples.
Biotech investors are AI-shy
“Everyone in the investment community realizes we’re at a transition point. Computing power is at the point where AI tools can be deployed, so there is a strong demand from the technology sector to fund companies that use machine and deep learning,” Alfa says. The biotech investors, generally speaking, are rather reluctant to go this route.
“There’s a sense that, ‘We tried this in the 1990s and it didn’t work,’” he jokes, but that’s actually close to the truth. There was a lot of hype about computational drug discovery that failed to reach fruition, so “investors are trigger-shy about getting back into it.”
Biotech companies, however, are starting to embrace AI, realizing it can unlock knowledge that until now has been hidden. To attract traditional biotech investors, avoid starting entirely from scratch. Instead, Singh says, “Use AI to build on existing knowledge. Then I might invest.”
Want to read more highlights? Click here to read more all of our Biotech Showcase coverage!
