Researchers at Qminers: From Bayesian Models to the Chemistry of Life

Science is one of the languages we speak at Qminers. We're not a research institute, but if you treated Qminers as a dataset and looked for common patterns across our careers, science would jump out fast. Many of us have published in peer reviewed journals, solved optimization problems in manufacturing processes, led seminars at the Academy of Sciences. Today we design strategies, build models, and write code, but the research mindset stays with us. It's a way of thinking. A sense of purpose. A feedback loop. A love of both the system and the detail.

Some of us still keep ties to academia. Mirek Rapčák still lectures, this fall he'll lead a seminar at the Institute of Theoretical Physics at the Academy of Sciences. Šárka Večeřová worked on developing a Bayesian model of moderated mediation that can capture complex relationships in hierarchical data with concurrent mediators and an ordinal outcome, for example when studying whether patients would recommend a hospital depending on its nursing care system. Vítek Kubelka finished his PhD in stochastic analysis while working and building his own project at the same time, and it's exactly that dynamic he wanted to carry into his next role. Filip Matzner is currently researching machine learning, focused on designing efficient recurrent neural networks for time series prediction, work he presented this year at the prestigious ICLR conference. And then there's Raman Samusevich. His story reads a bit like something out of Nature. Actually, quite literally.

When an ML model finds the scent of life in the oldest organisms

Raman Samusevich has been working since January this year in Qminers' deep learning team as an ML researcher. Before that he was a researcher at the Institute of Organic Chemistry and Biochemistry of the Czech Academy of Sciences (IOCB) and at CIIRC CTU.

"During covid I got fascinated by proteins. The precision and power packed into them felt almost alien. At the same time, breakthroughs in protein structure prediction were opening up a whole new world of understanding. It was the perfect moment to dive into bioML."

Out of intensive collaboration with IOCB came a research project bringing together biochemistry, machine learning, and a drive to get to the bottom of things, quite literally down to the molecular level. The goal was to find enzymes in DNA called terpene synthases (TPS). It might sound abstract, but the consequences are very concrete: TPS are responsible for most natural scents and flavors, including the ones used in perfumery, pharmaceuticals, and food production.

"When chemists can't make something efficiently, they steal it from nature. But nature doesn't produce these molecules on its own, enzymes do. We taught AI how to find these enzymes."

AI as a nose

Raman's ML model was able to read the presence of TPS out of DNA, and this led the researchers to the first functional discovery of these enzymes in Archaea, the oldest known domain of life. Biologists then inserted the DNA into yeast, and it actually started producing terpenoids. "It was like reconstructing the scent of something ancient for the first time in millions of years."

Raman and his supervisor Tomáš Pluskal built a series of talks based on the research, presented at Harvard, at DeepMind, and at the European Congress EUROBIOTECH. Their paper is currently in depth review at Nature Chemical Biology.

Side discoveries, side joys

While analyzing the data, Raman and his colleagues stumbled on other interesting findings too, for instance that viruses probably have a scent. (Yes, literally.) Their TPS turned out to be functional, opening up a new line of research connecting virology with chemical perception. At MIT he worked on optimizing known proteins, contributed to projects on catalytic DNA, and helped interpret spectra in mass spectrometry.

"It was a beautiful period. But then something happened that I'd never experienced in academia before. I met Filip Matzner, now CTO of Qminers, through one of his earlier projects. Using ML they designed a method to automate analysis of urban transport systems. Being able to push things through to production, fine tune the details, get feedback fast, it completely pulled me in. When Filip moved on to a new place but kept the same way of working, I went with him."

Back down to earth. Or into the office.

Raman is currently finishing his dissertation, handing the baton to his successor, and consulting on his last projects. He stays partly connected to academia, he holds a small position at IOCB, works externally at CTU, and keeps engaging with the scientific community. But his main energy now points elsewhere: to Qminers' deep learning team.

Trading and science turn out to be surprisingly close relatives. Raman's story is remarkable, but people with a research background are the norm here. And that's exactly what shapes a culture where analytical depth, genuine curiosity, and the drive to see things through come together naturally.