🌐 molecularmachinelearning.com · LinkedIn
We develop AI methods for drug discovery at the Eindhoven University of Technology, bringing together chemistry, biology and machine learning. We believe in open and reproducible research, so the code behind our papers is shared here. Some examples of what we do:
New deep learning approaches
- S4 for de novo design: chemical language modeling with structured state space models · Nature Communications
- JointMolecularModel: ML at the edge of chemical space, spotting out-of-distribution molecules · Nature Machine Intelligence
- ACER: generalizable protein–ligand co-folding · NeurIPS 2026
Tools
- ChemLint: MCP server that lets LLMs handle molecular data · J. Chem. Inf. Model.
- MolForge: reproducible molecular datasets at speed and scale · J. Chem. Inf. Model. (accepted)
- peptidy: peptide processing for machine learning · Bioinformatics Advances
Molecular benchmarks and systematic evaluations
- MoleculeACE: exposing the limitations of molecular ML with activity cliffs · J. Chem. Inf. Model.
- CycPepNet: comparing molecular representations and ML models on cyclic peptides · ChemRxiv
- MolFaith: assessing faithfulness of explainable AI in molecular deep learning · ChemRxiv
Active learning
- Traversing chemical space: active deep learning for low-data drug discovery · Nature Computational Science
- Self-driving labs: automated navigation of condensate phase behavior · Nature Communications
- Batch active learning: a systematic approach to batch active learning for material optimization · ChemRxiv
Molecular representations
- STRIPES: a physically interpretable symbolic language of protein–ligand interactions · ChemRxiv
- MATCHERS: a "chemistry-infused" representation bridging molecular graphs and strings · ChemRxiv
Browse all repositories for the code of our papers.
