EU-OPENSCREEN, in collaboration with the Society for Laboratory Automation and Screening (SLAS), is pleased to announce the winners of the Second Joint EU-OPENSCREEN / SLAS Machine Learning Challenge, revealed today at the SLAS 2026 International Conference & Exhibition in Boston, USA.

The challenge focused on predicting optical properties of small molecules, including UV/visible transmittance and fluorescence at bioassay-relevant wavelengths. High-quality datasets were generated with approximately 100,000 chemically diverse compounds as part of  EU-OPENSCREEN's bioprofiling activities. The participants developed innovative machine learning models to support assay design, compound triaging and hit selection in early drug discovery.

The competition attracted strong international interest, with 41 teams submitting 465 models. Applicants were from academia, non-profit organisations, and industry across Europe, Asia, and the Americas. This highlights the growing importance of Open Science, FAIR data and collaborative approaches in AI-informed chemical biology and drug discovery research.
    

Winning teams and solutions

Compound optical Transmittance prediction ML models

The first prize was awarded to Team “Yumiz” from the Ohue Laboratory, School of Computing, represented by Kairi Furui, Apakorn Kengkanna, and Koh Sakano. Their winning solution employed a multimodal molecular property prediction framework, integrating 1D, 2D, and 3D molecular representations. By combining multiple models — including gradient boosting, message-passing neural networks, and large pretrained molecular models — into a weighted ensemble, the team achieved robust and high-performing predictions on the blind test dataset.

Compound fluorescence prediction ML Models

The winning team in the fluorescence task was Team “Microsomes” from Meiji Pharmaceutical University, led by Wataru Miyahara, Kaito Inden, Yuma Iwashita, and Prof. Yoshihiro Uesawa. Their approach combined quantum-mechanical descriptors, large-scale cheminformatics features, and graph neural network embeddings using a sequential stacking strategy. Rigorous nested cross-validation and feature integration enabled highly accurate and generalisable fluorescence predictions.
    

Advancing open and collaborative machine learning

The Second Joint Machine Learning Challenge builds on the success of the first EU-OPENSCREEN / SLAS challenge and demonstrates how open, well-curated experimental datasets can accelerate the development of advanced machine learning methods for drug discovery. Full technical descriptions of the winning solutions will be published in SLAS Technology in 2026.

EU-OPENSCREEN and SLAS thank all participating teams for their contributions and look forward to future collaborations that further advance data-driven chemical biology research.