Artificial intelligence continues stirring things up in chemistry. To wit: YC-backed Cambridge, U.K.-based ReactWise is using AI to speed up chemical manufacture -- a key step in bringing new drugs to market.
Once a promising drug has been identified in the lab, pharma firms need to be able to produce much larger amounts of the material to run clinical trials. This is where ReactWise is offering to step in with its "AI copilot for chemical process optimization," which it says accelerates by 30x the standard trial-and-error-based process of figuring out the best method for making a drug.
"Making drugs is really like cooking," said co-founder and CEO Alexander Pomberger (pictured above left, with co-founder and CTO Daniel Wigh) in a call with TechCrunch. "You need to find the best recipe to make a drug with a high purity and a high yield."
The industry has for years relied upon what boils down to either trial-and-error or staff expertise for this "process development," he said. Adding automation into the mix offers a way to shrink how many iteration cycles are required to land on a solid recipe for manufacturing a drug.
The startup thinks it will be able to deliver "one shot prediction" -- where the AI will be able to "predict the ideal experiment" almost immediately, without the need for multiple iterations where data on each experiment is fed back in to further hone predictions -- in the near future ("in two years," is Pomberger's bet).
The startup's machine learning AI models can still deliver major savings by reducing how much iteration is required to get past this bit of the drug development chain.
"The inspiration for this was: I'm a chemist by training, I worked in Big Pharma, and I saw how tedious and trial-and-error driven the whole industry is," he said, adding that the business is essentially consolidating five years of academic research -- his doctorate focused on "the automation of chemical synthesis driven by robotic workflow and AI" -- into what he bills as "a simple software."
Underpinning ReactWise's product are "thousands" of reactions that the startup has performed in its labs in order to capture data-points to feed its AI-driven predictions. Pomberger says the startup used a "high throughput screening" method in its lab, which allowed it to screen 300 reactions at a time, enabling it to speed up the process of capturing all this training data for its AI.
"In pharma ... there are one or two handfuls of reactions, reaction types, that are used over and over again," he said. "What we are doing is we have a laboratory where we generate thousands of data points for these most relevant reactions, train foundational reactivity models on our side, and those models can fundamentally understand chemistry. And then when a client pharmaceutical company needs to develop a scalable process, they don't need to start from scratch."
The startup commenced this process of capturing reaction types to train its AIs last August, and Pomberger said it will be completed by the summer. It's working toward spanning 20,000 chemical data points to "cover the most important reactions".
"To get one single data point in a traditional manner it takes a chemist, typically, one to three days," he said, adding: "So this is really, we call it, expensive to evaluate data. It's very hard to get the single data points."
So far it's focused on manufacturing processes for "small molecule drugs," which Pomberger said can be used in medicines targeting all sorts of diseases. But he suggested that the technology could be applied in other disciplines, too, noting that the company is also working with two material manufacturers in polymer drug delivery development.
ReactWise's automation play also includes software that can interface with robotic lab equipment to further dial up precision manufacturing of drugs. Though, to be clear, it's purely focused on selling software; it's not a maker of robotic lab kit itself. Rather, it's adding another string to its bow in being able to offer to drive robotic lab equipment if its customers have such kit to hand.
The U.K. startup, which was founded in July 2024, has 12 pilot trials of its software up and running with pharma companies. Pomberger said they're expecting the first conversions -- into full-scale deployments of the subscription software -- later this year. And while it isn't revealing the names of all the firms it's working with yet, ReactWise says these trials include some Big Pharma players.
ReactWise is disclosing full details of its pre-seed raise, which totals $3.4 million, the startup exclusively told TechCrunch.
The figure includes previously disclosed backing from YC ($500,000) and an Innovate U.K. grant of close to £1.2 million (around $1.6 million). The rest of the funding (around $1.5 million) is coming from unnamed venture capitalists and angel investors, who ReactWise says are "committed to advancing AI-driven, sustainable pharmaceutical manufacturing."
While ReactWise is focusing, fairly narrowly, on a specific part of the drug development chain, Pomberger said acceleration here can make a meaningful difference in shrinking the time it takes to get new pharmaceuticals to patients.
"Let's look at a typical duration of a drug from start to launch: 10 to 12 years. Process development takes one to 1.5 to two years. And if we can basically speed up here the workflows -- reduce it by an average of 60% -- then we can get an idea of how much an effect it is," he noted.
Simultaneously, other startups are applying AI to different aspects of drug development, including identifying interesting chemicals in the first place, so there's likely to be compounding effects as more automation innovations get folded in.
But when it comes to drug manufacturing, specifically, Pomberger argues that ReactWise is ahead of the pack. "We were the first to actually tackle this," he said.
The startup competes with legacy software using statistical approaches, such as JMP. He also said that there are a few others applying AI to speed up drug manufacturing, but said that ReactWise's access to high-quality data-sets on chemical reactions gives it the competitive edge.
"We are the only ones that have the capability of, and that are currently generating, these high quality data sets in house," he said. "Most of our competitors, they provide the software. The clients are basically prompted with instructions based on the inputs.
"But, from our side of things, we offer these pretrained models -- and those are extremely powerful because they fundamentally understand chemistry. And the idea is then to really have a client just say: 'This is my reaction of interest, hit start, and we already give them process recommendations from the very first day, based on all the pre-work that we did in our laboratory. And that's something nobody else does at the moment."
