Dive Brief:
- Relay Therapeutics, a five-year-old biotech developing targeted cancer medicines, has acquired ZebiAI, a startup launched in 2019 to develop machine learning tools for drug discovery, the companies announced Friday.
- Relay will pay $85 million upfront for ZebiAI, split between $20 million in cash and $65 million in Relay shares. Shareholders in ZebiAI could receive as much as $85 million more in equity if certain milestones are met.
- ZebiAI specializes in using machine learning, a form of artificial intelligence that broadly describes self-learning computer algorithms, to assess data from screening target proteins against "libraries" of DNA-encoded molecules. Relay hopes ZebiAI's technology will help it more efficiently predict compounds with the kinds of chemical properties that make for a promising drug candidate.
Dive Insight:
The acquisition of ZebiAi is an unusual and early step for a company like Relay, which went public last July and only just began clinical testing of its lead drugs last year.
The biotech, which was launched in 2016 with backing from Third Rock Ventures, pitches itself as a new breed of biotech, blending experimental and computational techniques to discover and develop drugs better targeted at tough-to-reach proteins. Drawing on supercomputers built by backer D.E. Shaw, Relay designs its drug candidates by studying how those target proteins move, rather than using the static structures that have informed much of past decades of drug development.
Other biotech companies are trying something similar in mixing proven experimental methods with more novel approaches involving machine learning or other computing techniques. Enthusiasm for machine learning in drug discovery has led, like the advent of computer-aided drug design did beginning the 1980s, to criticism of excessive hype over its still unproven potential.
But Relay has something to show for its claims, having advanced two drug candidates into the clinic last year. A licensing deal with Roche in December, meanwhile, gave some external validation for the promise of its lead molecule, dubbed RLY-1971.
In an interview, Sanjiv Patel, Relay's CEO, described the company's progress and financial backing as a draw for smaller companies developing new computing techniques for drug discovery.
More and more founders began coming to Relay, said Patel, pitching the biotech on their platforms. Entering this year, Relay began taking those inquiries more seriously and grew interested in ZebiAI. In particular, a paper ZebiAI researchers published on their approach in the Journal of Medicinal Chemistry last year, drew Relay's eye, Patel said.
ZebiAI creates machine learning models that process data from screening through digital libraries of DNA-encoded small molecules — essentially massive databases of small molecules with DNA "barcodes" affixed to them. When researchers are interested in a particular protein associated with a disease, they can screen that protein against the library, generating an array of molecules that bind to the target. Sequencing the DNA then reveals the identity of those "hits."
But not all of those hits have properties that make them good candidates for drugs, or even easily chemically synthesized. ZebiAI's approach is to build a machine learning process by which active "drug-like" compounds can be more readily predicted.
Combining ZebiAI's platform with Relay's, Patel said in a statement, could also help reduce the time it takes to optimize a would-be drug candidate and potentially help increase the number of drug programs that can be advanced in parallel.
ZebiAI, which has partnership with X-Chem and Google Accelerated Science, has about 30 employees that will come over to Relay, now a company of about 170 staff.