Blog post image1

Molecule.one is a fast-growing company with a mission to teach computers chemistry so drugs can be discovered faster. Our team is expanding, and we’ve recently welcomed a number of new team members.

Molecule One-11 (1)Przemysław Pobrotyn joined the team in February 2021, and he sat down with us for an interview to share his experience with the company so far. This interview has been edited for clarity and brevity.

 

Tell us a little bit about your background and previous work experience. 

My education is in pure mathematics. I even started doing a Ph.D. in pure mathematics. Then I changed careers a few times before switching to machine learning. I was a software engineer in machine learning, at a large corporation here in Poland. While many of my colleagues here come from academic research, I’d say I’m coming more from the engineering side. 

 

What brought you to Molecule.one? 

The three main reasons I joined were, first of all, the unique challenge that the company is trying to solve. Everyone says they’re doing AI these days--it's not always obvious which companies offer an opportunity where you’ll actually be involved with state-of-the-art machine learning, versus those that only advertise that, and you end up doing some mundane logistic regressions. And here, because of the challenge, it was very certain that in order to solve that, you do have to be at the very forefront of the research to be able to tackle these sorts of challenges. And this was incredibly appealing. 

Secondly, because it's a startup, and it's a fairly early-stage startup, you have a much broader impact. In a large corporation, you're often kind of limited in the scope of your work. There's some inherent momentum to how quickly things can change. Here at Molecule.one, everything's much more fast-paced, and your impact is much broader. 

And finally, last but not least, I really like the mission of the company. Ultimately, if we achieve our end goal, we're doing something that's going to benefit society. Our mission statement is to help make medicines faster, which will have a positive impact on the world. The mission statement of the company is very clear: we're here not only to stay neutral, we're actually here to do good, which is very appealing.

What do you most appreciate about working at Molecule.one so far? 

I'd say that what I enjoy the most is that we have a lot of time to actually dig deeper into what we're doing. We have lots of uninterrupted “deep work” time, which is very precious. You don't get that a lot in larger organizations where you run from meeting to meeting, and you spend more time planning and reporting than actually pushing the boundary of what you're doing. 

At Molecule.one, I feel like we struck the right balance between planning what we set out to do, and actually having the freedom to do it in an uninterrupted way. That's what I enjoy the most, that we often get left to our own devices. On an average day, I might have one status update or daily meeting on my calendar, and then the rest of the day is free for me to just do my stuff. This is awesome!

 

What’s something you’ve learned at Molecule.one that you might not have learned somewhere else? 

I think what I'm actually still learning from my manager, is to be able to pivot quickly on to different topics. In a startup, it's very important to sometimes act on incomplete evidence, your gut feeling, and business needs. This means you have to be able to sometimes let go of an unfinished project because it's very likely that it's going to be a dead end. 

And it’s also important not to fall into the sunken-cost fallacy. It's much easier in a slower-paced environment, to turn back from a path once you’ve started and decided it’s not worth the time. Think about politics in academia or big corporate offices--you can’t admit that project wasn’t worth it to start with. Whereas in the startup, your time and resources are limited. You have to use them in the best way possible. Here, you don’t have to be afraid to let go of a project that maybe started off promising, but isn’t delivering on that promise. We have the freedom to let go and move on to the next thing. 

 

What challenges are you facing in machine learning? 

One thing we're looking at is in the field of machine learning. There's been this trend, actually for a number of years now, of pre-training large-scale models, and then fine-tuning them in downstream tasks. This is something we're looking at in the context of chemistry. 

There are obviously challenges and differences with how you do it. For instance, natural language is abundantly accessible on the web. Open chemical data isn't so abundantly accessible. There's that limitation of less data. And it's not entirely clear how to pre-train on that data. That’s the sort of challenge that we're looking at. 

Another thing we're looking at is, very broadly speaking, predicting reaction conditions. With a specific reaction, you want to choose the best set of conditions to carry out this reaction so that you maximize the yield of that reaction. Again, there are many different approaches you can take, and we’re investigating some of those approaches, which is very exciting. In some ways, this represents an extension of our MEGAN model, which proposes a number of ways to carry out specific reactions. 

 

What excites you the most about the future of the industry? 

There's a lot of talk in the industry about using AI, along with plenty of research. What I'd be excited to see is a prominent success story where there's a drug that's actually brought to market and that's actually saving lives, that's been developed thanks to these computational AI methods. It would be really exciting to have, say, a generative model first propose a molecule, and then our types of models propose how to synthesize a molecule, and then that actually undergoes human trials and ends up being in the market, thanks to technology.

On top of that, what if that kind of chemical molecule would have been from, let's say, the universe of all the possible molecules that previously were thought inaccessible for our synthesis methods. People are skeptical, so I think it would be particularly satisfying to see this come to fruition. 

Ready to participate in the future of drug development and synthesis planning? Molecule.one is hiring! Check out our current opportunities. 

Join Our Team

Author

Published

Author image

Weronika Kozyra

Leave a reply