• Period of stay: 22 to 29 April 2025

Report

My PhD research focuses on the development of novel machine learning methods for virtual screening, placing my work at the intersection of pharmacy and computer science. In computer science, major contributions are typically published in conference proceedings rather than in academic journals.

Bearing this in mind, I decided early on in my PhD that the core methodological paper of my thesis should be submitted to a leading computer science conference. The International Conference on Learning Representations (ICLR) is one of the three leading machine learning conferences, alongside NeurIPS and ICML. When I received the acceptance notification for my paper at ICLR 2025, I began planning my trip to Singapore, where I would present my work in the form of a poster. (By the way, you know you're on the right plane to a conference when half the passengers are carrying poster tubes in their hand luggage!)

ICLR is a massive and rapidly growing event, reflecting the immense global interest in machine learning. This year’s edition featured around 3,700 accepted papers and attracted approximately 10,000 attendees. Given its size, it was only possible to attend a small fraction of the presentations. Fortunately, the organisers did an excellent job of grouping posters and talks thematically. Despite each poster session including over 600 presenters, I could easily find the 20 or so posters related to chemistry and drug discovery. The conference also provided a mobile app that recommended relevant posters and talks based on individual research interests, helping to navigate the overwhelming number of sessions. Thanks to this organisation, it was surprisingly easy to connect with researchers in my field, despite the size of the venue.

For the presentations I was most interested in, I read the papers in advance. I can only recommend this strategy, as it allowed me to ask more specific questions during the sessions and engage in deeper, more meaningful discussions with the presenters. The most valuable part of the conference for me was presenting my own work. I received a lot of useful feedback and came away with several ideas for follow-up projects. It was also incredibly helpful to get a feel for the current trends in the field. Generative modelling is clearly one of the hottest topics in computer-aided drug design (CADD) at the moment. A significant proportion of the work I encountered involved flow-matching models conditioned on specific design tasks - an approach I’m considering exploring further in my own research.

After the three main conference days, ICLR continued with two days of workshops. These focused on specific research topics and mostly featured preprints and ongoing projects, giving a glimpse into what people are currently working on and how the field is evolving. I attended the workshop on generative modelling for biomolecular design. As in the main conference, many of the talks on CADD focused on flow matching in one form or another. While it was exciting to see how active and creative the community is, I also got the sense that the field is somewhat saturated with variations on the same core idea. Despite all the recent progress, it feels like we’re still waiting for a new paradigm - a real gamechanger that could steer the field in a new and promising direction.

In the evenings, there was also time to explore the city. I have to say that Singapore left a very positive impression on me. The city is incredibly clean, the public transport system is excellent, and the futuristic architecture is truly stunning. Everything was extremely well organised, both at the conference and in the city itself. I would like to thank my supervisor Thierry Langer, the Christian Doppler Laboratory for Molecular Informatics in the Biosciences, and the Vienna Doctoral School of Pharmaceutical, Nutritional and Sport Sciences for supporting this trip and the unforgettable experience!