Creating the AI-Powered Future of Chemistry
Polish Start-up Molecule.one Bridges the Gaps Between Chemistry, Technology and Business
Piotr Byrski and Paweł „Maxus” Włodarczyk-Pruszyński, who have known each other since high school and worked together for the last 10 years, created the first Molecule.one technology prototypes. They were joined by Paweł Łaskarzewski and Stanisław Jastrzębski as advisor, who added their business acumen and AI expertise to the team skillset. The founding team explains their motivation, current situation and next steps to develop the company.
CHEManager: What inspired you to found a company?
Piotr Byrski: While studying medicine, we saw many people suffering from untreatable diseases. That prompted us to ask ourselves two questions. Firstly, why designing a new drug must take so long. Secondly, what can we do — with our chemical and mathematical education — to solve this problem?
How did such a noble idea come to life?
Paweł Włodarczyk-Pruszyński: During our studies, together with Piotr, we were invited as consultants to a research group operating at the Institute of Organic Chemistry in Warsaw. A group of chemists worked there on an attempt to automate the synthesis of chemical compounds. We joined the team as consultants and contributed to the work resulting in the database to predict the methods of synthesis of organic compounds. After collaborating for some time, we decided to pursue our own idea inspired by the recent advances in the artificial intelligence field.
P. Byrski: Their solution provided good results, but unfortunately it was not sustainable to maintain a system based on hand-coded rules in the long run. Maxus and I thought that this was not a forward-looking approach. Chemistry is developing very dynamically, with new compounds and types of reactions appearing each day. We wanted our model not only to apply what it had in the set, but also to have some creativity. Combining math and chemistry skills allowed us to consider the perspectives of both users and software developers.
And what happened next?
Paweł Łaskarzewski: Then we met. After initial talks with pharmaceutical companies, I realized that the demand for such a solution is huge. Companies have been gathering information for years about the reactions they carry out in their laboratories and had a lot of unused data. We had an idea for a world-class AI tool that could predict if a chemical reaction could be carried out and how. From the beginning, our goal was to create a solution that could learn and be able to draw conclusions based on previously performed experiments.
After this analysis, I decided to help them from the business side, to raise capital, promote and support building high-class technology. In 2018, when the first investors appeared, we began building the system. Sunfish Partners, a VC fund based in Berlin, Germany, invested in our company. We have also filed a patent application in the US, where the largest number of pharmaceutical companies is located. We intend to open a company there.
What are your next steps in technology and business development?
P. Łaskarzewski: In terms of technology, we are constantly optimizing our machine-learning algorithm in order to improve the quality and the confidence measures of our predictions. Furthermore, we are increasing the performance of our solution. Although we can already analyze 10,000 compounds per hour, we are working to go beyond that. In addition, we aim to make our interface more convenient for our customers. In terms of business, we are planning to raise some funding this year. While continuing to work with our existing customers, we are also seeking to expand our customer base.
Molecule.one is supporting the efforts of scientists and researchers around the world to discover a drug against Covid-19.
P. Byrski: Yes. Recently, we have achieved a significant technological milestone, which enables us to plan and evaluate chemical synthesis for multiple compounds at once. This is useful for early stages of drug discovery, when lots of options need to be kept open. A couple of months ago, when the coronavirus crisis started to spread, we wondered how we could support the development of drugs against SARS-CoV-2. Given we are the only technology platform to perform synthesis planning for thousands of molecules per hour, we realized this was where we could bring value. Therefore, we decided to grant free access to our synthetic accessibility screening (SAS) capabilities to every team involved in developing potential treatments and cures for Covid-19.
Piotr Byrski is a Doctor of Medicine and holds a B.Sc. in Chemistry and Mathematics. A laureate of multiple national and international science competitions, he previously was a board member at Collegium Invisibile, an academic NGO.
Paweł “Maxus” Włodarczyk-Pruszyński, is a Doctor of Medicine and holds a B.Sc. in Chemistry and Mathematics. A silver medal laureate at the International Chemistry Olympiad, the self-taught software engineer has 7 years of experience in day-to-day programming.
Paweł Łaskarzewski exited two successful start-ups. The former CTO of Absolvent.pl: the 4th fastest growing start-up in CEE in 2016, built a mobile bank from scratch as CTO/COO. and has 20+ years of experience in ICT.
Stanisław Jastrzębski earned a Ph.D. from Jagiellonian University and was a postdoc at New York University. He published highly cited work with top scientists in deep learning (Yoshua Bengio, Kyunghyun Cho) and gained industrial experience with Google and Palantir. He has been supporting Molecule.one as start-up advisor and we'll be joining the company in September 2020.
Artificial Intelligence (AI) enables to speed up the process of chemical synthesis and, thus, the development of new drugs. The Molecule.one platform is a type of virtual laboratory where researchers can get a recipe for a given molecule, and then physically create it in the laboratory. In order to generate syntheses, it uses graph neural networks, adapted to the needs of chemistry. Since a chemical reaction is the transformation of one chemical compound into another, both can be saved as one graph.
If a scientist wants to learn how to create a new, never synthesized molecule, then Molecule.one’s artificial intelligence will cope with it. AI will look for similar reactions or chemical standards in the data, based on which it will generate a synthesis path — i. e. a recipe for how to do it.
The data is acquired from the US Patent Database. The patent register contains a huge amount of publicly available data from which millions of examples of chemical reactions can be obtained. The correct interpretation of information recorded this way is sometimes difficult and it takes a lot of time to pull out all the important elements. However, the start-up team managed to properly clean this data and today is automatically updating the database with new patents. Molecule.one is also able to clean up other data sources. Entering into a cooperation with a large pharmaceutical company will enable the start-up to connect their internal data sources, so that the algorithms can also take into account their experience and knowledge accumulated over the years.
Using this solution in the pharmaceutical industry can affect the lives and health of billions of people around the world. Automation of the synthesis process enables a much faster launch of new drugs on the market — maybe also a drug for the treatment of Covid-19.
All About the People
Molecule.one is a Polish start-up. Founded in 2016 in Warsaw, the company has created the fastest software platform for chemical synthesis based on state-of-the-art artificial intelligence (AI). The start-up wants to lay the groundwork for the automated future of organic chemistry, especially in the pharmaceutical industry.
- Proof of the ability to build the system and raise initial interest of 6 top pharmaceutical companies
- Angel investment allowing to accelerate development
- AI achieved first human-level results on a subset of reactions
- Roll-out of a beta version of the software to first users
- Closing of pre-seed venture capital Investment
- Deployment of first AI model to production
- Launch of product at TC Disrupt SF Startup Battlefield
- AI achieved state-of-the-art retrosynthesis results on a public bench-
- Synthetic Accessibility Score was used by researchers working on Covid-19 drug discovery
- First scientific papers published online
- Confirmation of trials with 10 top
- pharmaceutical companies
- Significant improvement of interpretability and robustness of AI models used
- First long-term partnership with a top pharmaceutical company
- Prediction of the cost of drugs synthesis using AI with similar accuracy as chemists
- Generation of reaction data and first AI-based predictions based on these data
- Equipment of platform with accurate reaction conditions and outcome prediction
Ul. Sienna 83/701