This Week In Cheminformatics: Issue #027

Strasbourg Summer School In Cheminformatics 2026, PP-MAPS, PegaPlus and a long list of papers.

Jun 29, 2026

Strasbourg Summer School in Chemoinformatics – 2026

The 10th edition of cheminformatics summer school in Strasbourg came to an on the 26th of June. Below are some of the more appropriate images that I found on my phone over the last week. Apologies to everyone who’s not in them.

Once again, many thanks to the organizers of the event and the co-ordinators of the Erasmus Mundus Master Program in ChEMoinformaticsPlus for their support.

Highlights

PegaPlus─Interactive Machine Learning by Human Observation for Efficient Clustering and Analysis of Structure–Activity Data

Clustering high-throughput screening hits into initial SAR series leaves medchemists with the tedious task of manually refining the groups to fix clustering / misclassifications. Fährrolfes et al. instead built a human-in-the-loop workflow to improve this. They start with a standard k-means++ assignment and present a flattened projection using a modified Stochastic Proximity Embedding algorithm. What makes this interesting is the feedback loop: when an expert manually drags a molecule to a different cluster to correct it, a modified Pegasos online Support Vector Machine instantly treats that reassignment as a hard constraint. It then incrementally updates the multi-class decision boundaries without retraining completely. The authors demonstrated that this interactive setup cuts the required manual refinement steps roughly in half across several benchmark datasets. Code can be found here. Good read !

PP-MAPS: Dynamic Pharmacophore Signatures of Protein–Peptide Interfaces from Molecular Dynamics Trajectories

Depenveiller et al. present a method, PP-MAPS, which extends dynophore analysis from small-molecule ligands to protein-peptide complexes. The practical obstacle is that peptides come encoded as part of the macromolecule in MD outputs, so most pharmacophore engines refuse to treat the peptide as a ligand. PP-MAPS solves this by enforcing the peptide at the end of each PDB frame and delegating feature detection to LigandScout or the open-source CDPKit. The output is a pharmacophore-type frequency matrix aggregated over the trajectory per residue, this makes comparisons across peptides and receptor variants possible. The CDPKit path runs end-to-end without a LigandScout license. Worth a read if you work on peptide therapeutics or need a compact way to summarize how a conformational ensemble fills a binding groove. Code can be found here.