Discover how companies are responsibly integrating AI in production. This invite-only event in SF will explore the intersection of technology and business. Find out how you can attend here.
Yoneda Labs, a Y Combinator-backed startup using AI to aid drug discoveries, announced it has raised $4 million in seed funding. The round was led by Khosla Ventures and included participation from 500 Emerging Europe, 468 Capital and YC. The capital will be used to acquire the robotic automation devices needed to run chemical reactions within Yoneda’s laboratory, which creates the training data for the startup’s model.
Founded by Michal Mgeladze-Arciuch, Daniel Vlasits and Jan Oboril, Yoneda Labs seeks to develop a foundation model for chemical manufacturing. “Whenever chemists need to make a new drug, we use AI to tell them how,” Mgeladze-Arciuch explained to VentureBeat. “This enables the creation of new drugs and makes chemical manufacturing faster and cheaper.”
“Machine learning and generative AI models have already begun to accelerate physics-oriented fields like aerospace engineering,” Jon Chu, a partner at Khosla Ventures remarked in a statement. “Chemistry will be no different and the team at Yoneda Labs has a novel approach to creating a foundation model for chemistry that could change the way chemicals are manufactured and improve the drug discovery process.”
Streamlining the drug framework process
Unsurprisingly, drug creation is not an easy process, especially when trying to put two molecules together to assemble a compound. The process is difficult because “you have to figure out how to make this reaction happen, how to make molecules actually stick together,” Mgeladze-Arciuch said. “You need to get the right temperature, solvent, base and all these different conditions that make the reactions happen.” Chemists spend an exorbitant amount of time simply doing trial and error to pair molecules, and Yoneda Labs believes it has an AI model that can speed things up.
VB Event
The AI Impact Tour – San Francisco
Join us as we navigate the complexities of responsibly integrating AI in business at the next stop of VB’s AI Impact Tour in San Francisco. Don’t miss out on the chance to gain insights from industry experts, network with like-minded innovators, and explore the future of GenAI with customer experiences and optimize business processes.
The startup pointed out that chemists working to synthesize drug compounds run trials in wet labs without any automation, simulation tools, or computational support. By operating its state-of-the-art wet lab, Yoneda Labs hopes to develop an AI that eliminates a major aspect of the pharmacology process, saving drugmakers time and resources.
Its main competitor is what’s listed in scientific literature. Mgeladze-Arciuch claimed that’s the current standard in the industry—chemists peruse books to find similar reactions that have been performed in the past, using that information to influence how they tackle the reaction they’re trying to make happen.
The race to 20,000 experiments
But it’s not there yet. It first needs to establish its training data, which requires running experiments. Yoneda Labs is shunning external data to educate its model to avoid bad data, an important aspect when working in science and medicine. Mgeladze-Arciuch stated it’s about quality versus quantity, so his company will produce the needed data in its web lab using robotic automation. “We’ll be able to generate 200 experiments every single day, which is roughly equivalent to what 20 chemists will be able to do. That’s how we will build our own proprietary data set that will then train the model on top of.”
It has a long way to go, too. The company estimates it has to run around 20,000 experiments to make its model commercially viable. It predicts it’ll complete this before the end of the year, and then Yoneda Labs could release its model.
Going after the small molecule
In such a broad industry, the company is focused on creating small-molecule compounds, which it says is a popular modality. “We are trying to make a model that will generalize across all the possible small molecules you can create. So far, we have tested our methods for the selection of experiments. So, one of the important parts of the pipeline for training this model is what experiments we will tell the robots to run. We’ve validated it on a couple of reactions that come from two very popular reaction classes in chemistry. And these reaction classes are very popular, particularly in medicinal chemistry. So at the point when you’re trying to discover new drugs.”
What are small molecules? Mgeladze-Arciuch explained that they make up a “significant portion” of the drugs available in the market today. In contrast, big molecules are used for protein-type drugs.
Yoneda Labs plans to have only one model to help chemists identify the organic reactions they want to happen and the ideal conditions needed. Ultimately, it hopes to be a “universal provider for any organic chemistry.” In fact, as Mgeladze-Arciuch put it, “For the creation of any small molecule, we want to be able to tell the chemists how to make these molecules. Effectively, we want to be like OpenAI for chemistry. Whenever a chemist needs to create an organic small molecule, they put this into our model, and our model will give them the recipe for making it.”
