Artificial intelligence continues to inspire things in chemistry. Witness: Cambridge University supported by the UK YC reaction AI is being used to speed up chemical production, which is a key step in bringing new drugs to the market.
Once promising drugs are identified in the laboratory, pharmaceutical companies need to be able to produce large amounts of materials for clinical trials. Here, ReactWise provides interventions to intervene in its “chemical process optimization AI copilot,” which it says accelerates 30 times, based on standard trials and error processes to figure out the best way to make drugs.
“Drug use is really like cooking,” said Alexander Pomberger, co-founder and CEO (pictured above left, with co-founder and CTO Daniel Wigh) on a phone call with TechCrunch. “You need to find the best recipes to make medicines with high purity and high yields.”
For years, the industry has relied on trial or employee expertise that comes down to this “process development.” Adding automation to the mixture provides a way to narrow down how many iteration cycles are required to land on solid formulations that make the drug.
The startup believes it will be able to provide “one lens prediction” (AI) almost immediately “predicting the ideal experiment” without multiple iterations, in which case the data in each experiment is fed back to further hone the prediction (in the near future (“in two years,” is Pomberger’s bet).
The startup’s machine-learning AI model can still save a lot of savings by reducing iterations to go beyond the drug development chain.
Cut through boring
“It’s the inspiration for this: I’m a trained chemist, working at a big pharmaceutical company, and I’ve seen the tedious and trial-and-error throughout the industry,” he said. He added that the business essentially cemented his PhD over five years of academic research – his PhD focused on “automated automation of robotic workflows and automation of AI, with robotic workflows and AI as actions –
The foundation of ReactWise is based on the “thousands of thousands” of reactions the startup performs in its labs to capture data points to supply its AI-driven predictions. Pomberger said the startup used a “high-throughput screening” method in its labs that enables it to screen 300 reactions at a time, allowing it to speed up the process of capturing all training data from its AI.
“In pharma, one or two reactions are used over and over again, reaction types,” he said. “What we are doing is that we have a lab where we generate thousands of data points for these most relevant reactions, training fundamental reactive models around us that fundamentally understand chemical reactions. Then, when a customer pharma company needs to develop scalable processes, they don’t have to start from scratch.”
The startup began the process of capturing reaction types, training its AIS last August, which Pomberger said will be done in the summer. It is working to span 20,000 chemical data points to “cover the most important reactions.”
“To get a data point in the traditional way, it usually takes one to three days,” he said, adding: “It’s really, we call it expensive. It’s hard to get a single data point.”
So far, its focus is on the manufacturing process for “small molecule drugs”, which Pomberger says can be used for drugs targeting a variety of diseases. But he suggested that the technology could also be applied in other disciplines, noting that the company is also working with two material manufacturers for polymer drug delivery development.
ReactWise’s automated game also includes software that can be connected to robotic lab devices to further dial the drug precisely manufactured. But, it is clear that it is purely focused on selling software; it is not the manufacturer of the robot lab kit itself. Instead, if its customers can use such a toolkit, it is providing another string for the bow.
The UK startup, founded in July 2024, conducted 12 pilot trials of its software with pharmaceutical companies. Pomberger said they expect a first conversion later this year – fully deploying subscription software. While it did not name all the companies it worked with, ReactWise said the trials included some large pharmaceutical companies.
Pre-planting funds
The startup exclusively told TechCrunch that ReactWise is revealing full details of its pre-planting prepayment, totaling $3.4 million.
This figure includes support for previously disclosed YC ($500,000) Innovate UK Grant Nearly £1.2 million (about $1.6 million). The remaining funds (about $1.5 million) come from unnamed venture capitalists and angel investors who said they are “committed to advancing AI-driven, sustainable pharmaceutical manufacturing.”
While ReactWise’s concentration is quite narrow, on a specific part of the drug development chain, Pomberger said acceleration can make meaningful changes in the time it takes for patients to obtain new drugs.
“Let’s look at the typical drug duration 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 the workflow here (average reduction of 60%), then we can see how effective it is,” he said.
Meanwhile, other startups are Apply AI to different aspects of drug developmentincluding identifying interesting chemicals first, so as more automation innovations are folded, more complex effects may be produced.
But when it comes to drug manufacturing, Pomberger believes ReactWise is in packaging. “We are the first ones to really solve this problem,” he said.
The startup uses statistical methods, such as JMP, to compete with old software. He also said there are other applications of AI to speed up drug manufacturing, but he said ReactWise gains access to high-quality datasets on chemical reactions to give it a competitive advantage.
“We are the only ones with the capability to these high-quality data sets and the capabilities currently being generated,” he said. “Most of our competitors offer the software. Basically, it’s prompting customers based on the instructions entered.
“But, from our perspective, we offer these validated models – these models are very powerful because they fundamentally understand chemistry. Then the idea is to really get the client to just say, “This is the reaction I’m interested in, hit rate, and we’ve already advised them on the process based on all the work we’ve done in the lab.” This is something no one else is doing at the moment. ”
