Pharmaceutical companies around the world are currently working hard to discover potential treatments and cures for COVID-19. One big challenge is the long time that is usually needed to develop a new drug. At this point, the Polish startup Molecule.one – participant of this year’s startup program 5-HT X-linker – wants to help by granting scientists free access to its platform. In this interview, CEO and co-founder Piotr Byrski M.D., explains how their AI platform makes it possible to accelerate the process of chemical synthesis, driven by the idea to make medicines faster.
What is the vision of Molecule.one?
We want to lay the groundwork for the automated future of organic chemistry, especially in pharmaceutical industry. With artificial intelligence, we want to speed up the process of chemical synthesis and, thus, the development of new drugs and other molecules.
What are the biggest challenges when it comes to the current methods of chemical synthesis in pharmaceutical industry?
Chemical synthesis is one of the bottlenecks in drug discovery. It is one of the reasons why the discovery of a new drug needs as much as 12 years in average. Making a wrong decision can cost you several weeks, and this happens multiple times throughout a single drug discovery pipeline. Even for the synthesis of modestly complex molecules, there is an enormous number of possibilities. In order to choose the materials, reactions and conditions which will deliver the desired results, you need to plan the process very well and consider a large number of datapoints. This requires a lot of time and effort.
How can Molecule.one improve the process of chemical synthesis?
The users of our platform benefit from significant time and cost savings. An individual chemist can maybe take 10 or 100 previously performed reactions into account when planning a synthesis – our algorithm can look at millions of reactions and evaluate which of these are going to work in a specific case. The technological solution of Molecule.one enables chemists to spend their time thinking about more complex problems instead of manually going through publications about chemical synthesis. Furthermore, the algorithm allows our clients to set priorities, depending on which factors they want to optimize: are they looking for the cheapest way to synthesize a new molecule, or rather for the fastest way? Do they want to produce small or large quantities? Finally, our solution proposes relevant reactions, scores them and assembles them into full synthesis pathways – all within a few minutes.
What is the technology behind your solution?
There are multiple pieces of technology at work. First of all, we use databases that contain information about previous reactions, commercially available starting materials or known molecules. After extracting the knowledge from these databases, for example by natural language processing, we use a machine learning approach to find answers to the questions of our clients. Last but not least, we aim to be as convenient as possible for the end user. Therefore, we work together with our customers to constantly validate our assumptions and to rapidly improve the way we present the data to our customers.
How do know that your system has identified the best solution for the synthesis of a specific molecule?
It is very important for us to make sure that our algorithm works correctly. Therefore, we apply various lines of verification. First, we use classical machine learning approaches to validate if we are actually able to generalize knowledge from datasets. Then, there is a chemist’s eye verification: we ask chemists to assess whether the proposed reactions are possible or not. Sometimes, they apply our algorithm to a problem that they already know the solution to (without the algorithm having any explicit access to this information). This is a good way to evaluate the reliability of our algorithm. As a third line of verification, we will also conduct laboratory experiments in collaboration with partners.
At the moment, scientists around the world are striving to discover a drug against the new corona virus. How does Molecule.one support these efforts?
Recently, we have achieved a significant technological development which enables us to plan and evaluate chemical synthesis for multiple compounds at once. This is useful for early stages of drug discovery when we need to keep several options open. A couple of weeks ago, when the corona crisis started to intensify in various countries, we wondered how we could support the development of drugs against SARS-CoV-2. We believe we are the only technology platform able to perform synthesis planning for thousands of molecules per hour. Therefore, we have decided to grant free access to our synthetic accessibility screening (SAS) capabilities for every team involved in developing potential treatments and cures for COVID-19. You can simply write us an e-mail, tell us about your idea, and then use our solution to assess thousands of molecules per hour in terms of how easy they are to synthesize.
How did the idea for Molecule.one develop?
The idea arose from experiences of myself and my co-founder Maxus (Paweł Włodarczyk-Pruszyński) who studied chemistry, mathematics and medicine together with me. During our time at university, we worked in laboratories, did consulting for academic teams, and were also active in the chemical industry. Soon we realized that academic approaches were often problematic: academic teams often tend to hyperfocus on particular technologies, and only afterwards they try to find industrial applications for their discoveries. We decided to work the other way round: Instead of commercializing research, we focus first on solving the problems that the industry actually has. In November 2016, we founded our company and started developing our technology. At the moment, we are a team of 9 people and will hopefully continue to grow fast.
What are your next steps in technology and business development?
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.
This year, you were part of our startup bootcamp X-Linker. What was your experience with the program – and what were your biggest learnings?
It was a great opportunity to get in touch with industry peers and with other startups doing excellent work. The most important thing for us was getting multiple perspectives on what we do. Especially the mentoring sessions were very valuable to us: we practiced how to explain our idea to different kinds of people; we identified risks and learned how to address them. The networking part was also a big advantage of the event, as it gave us the chance to lead interesting conversations, e.g. with BASF.