Yiannis Galdadas

Foldseek-Interface reveals a protein interface universe far from complete 1 Foldseek-Interface introduces an interface-specific structural encoding: it extracts contacting residues from both partners (Cα distance ≤10 Å; ≥4 residues per chain), renumbers them in N-to-C order within each chain, concatenates both sides, then converts the interface into a 3Di “sequence” for fast alignment. 2 By turning 3D interface comparison into sequence-like alignment, Foldseek-Interface reaches near state-of-the-art accuracy while being up to 230× faster than iAlign on benchmark sets spanning order–order and disorder–order interactions (with disorder-mediated cases remaining harder for both methods). 3 The work adds an interface-scale clustering framework (Foldseek-Interface-Cluster) calibrated using interface TM-score thresholds; benchmarks support a match cutoff around TM-score 0.3, while full-PDB clustering favored a stricter 0.4 to reduce spurious merges. 4 Applied to all biological assemblies in the PDB, the pipeline clusters 3,121,961 dimers into 77,167 interface clusters (via an intermediate redundancy reduction to 189,830 non-redundant dimers). The full run completed in ~6 days (GPU + 128-core CPU), making interactome-scale interface mapping computationally feasible. 5 The resulting “interface cluster” resource shows heavy skew: more than half of clusters contain only ordered interfaces, contrasting with expectations that many eukaryotic PPIs are disorder-mediated—highlighting a strong experimental bias in the PDB toward ordered binding modes. 6 Interface discovery appears to keep growing even as fold discovery plateaus: newly observed interface clusters continue to accumulate over time, suggesting the experimentally resolved interface universe is still incomplete. 7 By focusing on local interface geometry rather than full-chain folds, the resource reveals rare but informative cases where similar interfaces are carried by different folds (candidate convergent interface evolution). Only ~5% of multi-annotated clusters include more than one distinct CATH domain pair, but many of these span different CATH classes and tend to be helical at the interface. 8 Evolutionary stratification using taxonomic lowest common ancestor (LCA) suggests: ~4.4% of clusters are “old” (present across Eukaryota/Bacteria/Archaea), ~50% are species-specific (with many human-only). “Young” clusters are more enriched in disorder-mediated interfaces (21% vs 13% in old), aligning with increased regulatory/signaling complexity in multicellularity. 9 The interface clusters enable systematic mining of pathogen mimicry: after stringent filtering, 56 candidate clusters contain both human–human and human–pathogen interfaces (mostly viral). Viral mimicry clusters show higher disorder-mediated binding, consistent with short linear motifs as compact, rapidly evolving hijacking strategies. 10 Foldseek-Interface also scales to predicted complexes: searching ~21k predicted human binary complexes (HumanPPI) yields 10,040 clusters, including 1,780 putatively novel clusters with no similarity to any PDB interface representative (mostly singletons). Manual inspection flags artifacts but still identifies plausible novel interfaces (e.g., translation/ribosome-associated, cilia trafficking-related, and IDR-wrapping interactions) for experimental follow-up. 💻Code: github.com/steineggerlab/fol… ; github.com/soedinglab/MMseqs… ; github.com/rachelse/bfvd-web ; github.com/steineggerlab/fol… 📜Paper: biorxiv.org/content/10.64898… #computationalbiology #structuralbioinformatics #proteininterfaces #PPI #Foldseek #AlphaFold #disorder #evolution #pathogens #webserver

A Foundation Chemical Language Model for Comprehensive Fragment-Based Drug Discovery 1. FragAtlas-62M is a groundbreaking chemical language model specifically designed for fragment-based drug discovery. It is trained on the largest fragment dataset to date, comprising over 62 million molecules from the ZINC-22 database. This model achieves an impressive 99.90% chemical validity in generated fragments, making it a powerful tool for medicinal chemistry. 2. The model not only maintains a high coverage of known ZINC fragments (53.55%) but also generates 22.04% novel structures with practical relevance. This balance between rediscovery and novelty is crucial for fragment-based drug discovery, as it ensures both the reliability of known fragments and the potential for new discoveries. 3. FragAtlas-62M is built on a GPT-2 architecture with 42.7M parameters. It uses a 128-token context window and is trained using HuggingFace Transformers. The model's architecture and training methodology are optimized for fragment-level SMILES modeling, ensuring efficient and high-throughput generation capabilities. 4. The model's performance is validated across 12 molecular descriptors and three fingerprint methods. The generated fragments closely match the training distribution, with all effect sizes being less than 0.4. This indicates that the model generalizes well and maintains the key properties of the training set without systematic bias. 5. FragAtlas-62M demonstrates substantial overlap in chemotype space between novel and rediscovered molecules, as shown by t-SNE visualizations and distance analysis. The distance ratios (NR/NN and NR/RR) are consistently near 1.0 across all three fingerprint types, indicating minimal distributional shifts. 6. The model is released with training code, preprocessed data, documentation, and model weights, making it accessible for further research and practical applications. This open release lowers the barrier to entry for groups with modest computational resources and encourages rapid follow-up work and experimental validation. 7. While FragAtlas-62M is a significant advancement, it has limitations. It does not explicitly model stereochemistry, geometric relationships, or fragment-to-fragment connectivity rules. Future work should focus on integrating conditional controls, structural information, and methods for automated molecule construction to broaden its practical applications. 📜Paper: arxiv.org/abs/2509.19586 #FragAtlas62M #ChemicalLanguageModel #FragmentBasedDrugDiscovery #MedicinalChemistry #AIinDrugDiscovery #OpenSource #Research