Onepot AI raises $13M to help make chemical drug creation easier

For Daniil Boiko and Andrei Tyrin, the idea for Onepot AI came from the same frustration.  

“The best ideas successful cause find were often blocked not by biology, but by synthesis,” Boiko told TechCrunch. Synthesis is the creation of caller molecules by utilizing chemic reactions. It’s like a look oregon Lego pieces, wherever tiny pieces, ingredients, molecules, travel unneurotic to signifier a wider puzzle picture, a nutrient dish, a larger molecule.  

As 1 mightiness expect, it’s quite hard to make those tiny molecules that spell connected to physique bigger ones.  

For Boiko, a Ph.D. candidate studying instrumentality learning successful chemistry at Carnegie Mellon (he received his bachelor’s and master’s successful integrated chemistry from a university in Russia), that meant realizing that cause hunters — the scientists who oversee cause find and improvement — were skipping promising ideas conscionable because the chemical molecules to create the drugs seemed excessively hard to make.  

“The compounds ne'er adjacent got a accidental to beryllium tested,” Boiko told TechCrunch.  

For Tyrin (who received his bachelor’s in machine subject astatine MIT), his clip working on cause find computational pipelines made him recognize however behind the world of drug discovery was. “The models could make ideas in hours, but it could instrumentality months for the lab to drawback up,” helium told TechCrunch.  

“We some saw that the world was throwing wealth into molecular plan and astir ignoring the harder occupation of really making the molecules,” Boiko said. But there was a geopolitical space too, he continued, planetary proviso chains are becoming vulnerable, and the U.S. is entering a commercialized warfare and innovative competition, erstwhile again, with China.  

“It was clear,” Boiko said. “Small molecule synthesis needed to beryllium rebuilt from the ground up in the United States.” 

Boiko and Tyrin came together to create Onepot, a institution that is location to the small-molecule synthesis laboratory POT-1. They besides built Phil, an AI integrated chemist, to assistance tally experimental investigation to increase the process of compound synthesis for their aboriginal commercialized partners. Those partners are biotech and pharma companies that are currently trying out their technology.  

On Wednesday, the company came retired of stealth with $13 million of funding, including pre-seed money, and a seed round led by Fifty Years.  

“Currently, pharma and biotech companies either physique full teams of chemists in-house or enactment with declaration probe organizations overseas,” Tryin said of the process for molecular synthesis. Human chemists can spend months of probe to create even a azygous compound, at a cost of thousands of dollars.  

It involves a batch of proceedings and error — studying assorted compounds, collecting data on biological activity, how the drug moves through the body, toxicology reports, and coming up with what to experimentation with next. “The main limiting origin present is not testing these compounds, but making them in the first place,” Tyrin continued. “We purpose to compress this down to days.”  

Tyrin said the product is beauteous straightforward. Onepot has a catalogue of molecules it tin make. Clients take which compounds it want and then Onepot’s technology volition synthesize the molecules and then ship them to the customer so the customer tin use them in their ain experiments. (They vessel carnal products either arsenic adust compounds of solutions in plates or vials).  

The backend of the product is wherever Boiko and Tyrin person fun, dissecting the problems of chemic synthesis to find out which combinations of molecules enactment together. They built a laboratory where they are letting LLM agents access these alleged molecule recipes for training so the agents tin also find retired what works and doesn’t in compound building.  

“When executing experiments in the lab, we seizure each azygous item that goes into the process,” Tyrin said — that means tracing somesthesia and, essentially, the ingredients that were added to a substance to make compounds. “No accusation is lost, which makes experiences reproducible adjacent if idiosyncratic decides to run them successful 10 years from now.”  

This also means their agents generate hypotheses from real-world experiments alternatively than literature data, often mined from the internet.  

Boiko called the fundraising process “hectic,” and said they met their lead investor through an intro. “What was expected to beryllium a abbreviated gathering turned into a multi-hour whiteboard league astir industrializing synthesis,” Boiko said. Others in the round see Khosla Ventures, Speedinvest, OpenAI co-founder Wojciech Zaremba and Google’s Chief Scientist Jeff Dean.  

The fresh superior volition beryllium utilized to physique a 2nd laboratory successful San Francisco so the team tin instrumentality connected much customers. It volition besides expand the team and its compound find engine. On the service side, Boiko and Tyrin look at WuXi AppTec and Enamine arsenic competitors.  

Overall, Boiko and Tyrin anticipation to marque cause find astatine slightest 2 times faster and hopefully change the perception of what is imaginable aft 1 taps into the “weird” chemistry that scientists erstwhile thought was disconnected limits.  

“You’re not conscionable speeding up cause discovery, you’re expanding the design abstraction for what drugs and materials can be,” Boiko said. “That cause that we haven’t discovered yet, mightiness beryllium out there, waiting for america to find it.”