Our preprint “Explainable ML force-field for evaluating protein–ligand binding energy using SO3LR” is now out on ChemRxiv!
In our SO3LR-SF project, we introduce a fast and explainable scoring function for protein–ligand binding built on the pretrained SO3LR machine-learned force field. It brings near-quantum-mechanical accuracy to structure-based drug design at a fraction of the cost, and shows why a ligand scores the way it does. The code is available on GitHub.
Many thanks to our collaborators Sergio Suárez-Dou, Adil Kabylda, and Alexandre Tkatchenko (University of Luxembourg) for working together on this project.
Read more
This year, Volkamerlab co-organized the ELLIS summer school for Trustworthy & Responsible AI in Drug Discovery. Almost everyone from the chair was involved: Loulwah presented her first poster, “Assessing Robustness of Few-Shot Learning for Low-Data Toxicity Prediction.” Michael, Joschka, and Lisa supported Andrea in delivering a workshop titled “Establishing Hands-On Talktorials on Trustworthy AI for Drug Design.” Floriane, Yanyuan, Max, Ben, and Fatemeh volunteered as helpers throughout the event.
Overall, the summer school was a great success, featuring many inspiring speakers and numerous scientific highlights. It also provided an excellent opportunity for networking, socializing, and having fun. In particular, the social events, including the tour of the Völklinger Hütte and the barbecue, were a highlight of the week.
Read more
SBI² Premier European Meeting in Cambridge
In July, I had the opportunity to attend the Premier European Meeting of the Society of Biomolecular Imaging and Informatics (SBI²) in Cambridge, UK — the first-ever European meeting organised by SBI². The meeting took place on 7–8 July 2026, bringing together researchers, scientists and experts working at the intersection of biomolecular imaging, high-content screening and informatics.
The two-day programme combined educational sessions, keynote presentations, scientific talks and poster discussions, with a strong focus on how imaging and informatics can be integrated to extract meaningful biological information from high-content data. Discussions around high-content screening in drug discovery, image analysis and the use of quantitative morphological information highlighted both the opportunities and the challenges of translating imaging data into actionable insights.
Read more
We are happy to announce that the preprint of our paper “A Comprehensive Evaluation of Machine Learning Pipelines for Toxic Endpoint Prediction” is finally live on Research Square: https://www.researchsquare.com/article/rs-10470980/v1. The paper presents the first comprehensive benchmark of ML models, molecular representations, and DR methods for hormone receptor activity assays in ToxCast. In contrast to prior work, we systematically evaluate complete modelling pipelines with rigorous hyperparameter tuning and chemically-informed data splitting, supported by extensive statistical analysis. Our main finding is that simple approaches often outperform more complex alternatives. In fact, the combination of random forest with physicochemical properties and no dimension reduction emerged as the strongest and most consistent performer. Notably, it also outperforms deep neural networks and learned compound representations. Moreover, our analysis revealed that activity cliffs remain one of the major challenges for machine learning-based toxicity prediction.
Read more
As part of the 10th Strasbourg Summer School in Chemoinformatics (CS3-2026)
(CS3-2026 conference website), Andrea, Walid, and Ana represented our team at the University of Strasbourg from June 22 to 26, 2026. The summer school brought together students, early-career researchers, and experienced scientists to discuss recent developments in chemoinformatics, artificial intelligence in chemistry, chemical space exploration, QSAR modeling, and computer-aided drug design.
Andrea delivered a talk titled “Data-Driven Exploration of Kinase Inhibitor Space: AI-Assisted, Structure-Based, and Fragment-Guided Discovery”.
(talk abstract).
Read more
New Book Chapter: Morphological Data Analysis for Predictive Modelling
We are happy to share that a new protocol has been published in Methods in Molecular Biology (MIMB, Volume 2989) titled
“Morphological Data Analysis: From Descriptor Development to Predictive Modelling.”
The chapter is based on work carried out during the first year of my PhD project and focuses on the computational analysis of morphological fingerprints derived from Cell Painting assays. These fingerprints capture quantitative information about cellular morphology, texture, and organelle organization, and can be used for downstream applications such as compound similarity search and activity prediction using machine learning methods.
Read more
This year’s ELLIS workshop ML4Molecules was held in conjunction with EurIPS
in Copenhagen.
Michael Backenköhler and Joschka Gross presented two posters on recent work in machine learning using ChEMBL data.
The first poster presents the ideas laid out in “Assay-Based Machine Learning: Rethinking Evaluation in Drug Discovery”.
Public datasets are typically aggregates of experimental data originating from multiple laboratories. As a result, inconsistencies in the data are unavoidable. In this work, we investigate how conclusions drawn from standard machine learning workflows change when the assumption of measurement consistency is relaxed.
Read more
We are happy to announce that the preprint of our paper “The Trustworthiness Landscape in Machine Learning: A Conceptual Guide with Applications in Medicine” is finally live on Zenodo: https://zenodo.org/records/17591544. It explains more than 17 concepts related to the trustworthiness of machine learning (ML), including, e.g., reliability, robustness, fairness, interpretability/explainability, security and privacy, and explores how these concepts connect and conflict.
It is targeted at anyone who wants to use ML responsibly, and especially researchers who encounter ML-based science in their daily work (not necessarily as developers). No extensive ML background is expected.
Read more
Last week, the CytoData Hackathon in Berlin-Buch brought together researchers from different backgrounds to work on challenges in bio-image–based profiling, offering a great space for collaboration and exchange.
This was followed by the #CytoData2025 Symposium, where I presented ongoing work titled “Uncovering Compound Mode of Action via Hierarchical Clustering of Structural and Morphological Profiles” as part of the Morphology-based Endocrine Disruptors Screening project. The project aims to predict endocrine disruption by combining Cell Painting profiles with cheminformatics methods.
Read more
For this year’s group retreat, we kicked things off with a fun bowling outing that brought out some friendly competition. The evening continued with pizza and drinks, giving everyone a chance to catch up outside of work. It was a great opportunity to unwind together and strengthen connections as a group.
Read more
