Chemist Arina Shelashen researches AI-assisted drug development. She acquired her computer science expertise as a scholarship holder at the Konrad Zuse Schools of Excellence in AI, a programme run by the DAAD (German Academic Exchange Service). She now works as a doctoral researcher at the Institute of Bioinformatics at the University of Leipzig. Her goal: to determine which molecules proposed by artificial intelligence can actually be synthesised in practice.

Building molecules with AI is already possible today — though the programmes sometimes overshoot the mark. “In the simulation, everything looks fine. You've defined what properties a particular drug should have, for example,” explains Arina Shelashen. “But then it turns out that the substance can't be produced the way the AI envisioned it. Chemists can't synthesise it, because it's unstable, too expensive, or requires too many synthesis steps.”

As a doctoral researcher at the Institute of Bioinformatics at the University of Leipzig, Shelashen is working to close exactly this gap. The goal is to develop an AI-based software that can predict whether a hypothetical molecule is actually feasible to produce — and not just whether it can be made, but, as a next step, precisely how. “Which synthesis route is viable, and how efficient is it? That's highly complex, but these are exactly the kinds of tasks that artificial intelligence can handle extremely well.” Currently, this is limited to a single laboratory’s setup and has not yet been rolled out across the board.

Out of the Black Box

Such AI-based approaches are nothing new. However, they operate as black boxes, Arina Shelashen explains. “You can't trace how they arrived at their assessment.” She sees her task as adding mathematically grounded explainability to these prediction tools.

“We feed the system with all the reactions a laboratory can carry out — all the reagents it already has or can obtain — and derive a synthesis route for a target molecule from that.” Simulating this on the basis of all possible chemical reactions and compounds would be combinatorially unfeasible — “far too many possibilities,” she says. “But AI narrows down the search space, and mathematics verifies the result. That's our approach.”

Limited Options in Belarus

Shelashen grew up in Belarus. Mathematics and computer science always interested her, but after school she initially chose to study chemistry. After completing her bachelor's degree at Belarusian State University in Minsk, she worked for a pharmaceutical company before beginning to teach herself computer science on the side. “It was around 2021, and it was an exciting time. Large language models were slowly gaining traction. After work, I was sitting in front of my computer almost every evening, taking online courses in computer science.” After a few weeks, she realised: “This is very relevant to my work as a chemist — I really want to combine the two.”

Finding a suitable master's programme was not easy. “In Belarus, I had very few options. The academic system there is still very traditional, and interdisciplinary approaches tend to be viewed with scepticism.” Shelashen eventually found what she was looking for at TU Dresden (Dresden University of Technology). “They offered the master's programme in Computational Modeling and Simulation, with a direct connection to the life sciences. It was a perfect fit.” What makes it distinctive: it allows people without a formal background in computer science — including chemists — to study the subject. “At most universities, you need a degree in mathematics or computer science to enrol in a master's programme at a faculty of computer science. At TU Dresden, you have to demonstrate that you can code, but entry is possible without the formal education.”

There she learned about the opportunity to receive funding through the DAAD programme Konrad Zuse Schools of Excellence in AI. She ultimately secured a grant from the Zuse School SECAI, based at the Dresden and Leipzig sites. “That was extremely helpful — not just because of the financial support,“ she says. “The mentoring was also very important to me. I'm still in touch with my supervisors to this day.”

Cheminformatics — A Discipline with a Future

A typical working day for Shelashen is spent at the computer. She is no longer involved with chemical processes in the laboratory. The data she needs comes from the chemists on collaborative teams, or she attends workshops and conferences. “It's very important to always stay up to date — especially in a research field as specialised as mine.”

Although she works at the Institute of Bioinformatics, Shelashen describes herself as a cheminformatician. The distinction matters: while bioinformatics deals with large molecules — DNA, RNA, proteins — Shelashen and her team focus on molecules that are far smaller, “sometimes by a factor of 1,000. The fact that we work alongside the bioinformaticians in Leipzig is simply because our discipline isn't quite as well established yet.”

“I feel at home in Germany”

If Arina Shelashen has her way, that could soon change. She sees the opportunity to develop “something that truly matters” through her research, as she puts it. She has already been looking into positions — in the field of computational simulation for chemistry, for instance. “Those roles are rare and highly sought after. A doctoral background in this area is absolutely mandatory.” 

She can very well imagine staying in Germany. “I feel at home here. Many Belarusians have gone to Poland, which is much closer to us culturally and linguistically. But I feel more at ease in Germany. And safer — in more ways than one.”

Klaus Lüber (7 July 2026)