About PLINDER
Protein-ligand interactions are foundational to the understanding of science and discovery of therapies. However, to this date, no large, high quality datasets with real-life relevant evaluations exist.
PLINDER is an academic-industry collaboration to address this, driven by VantAI, NVIDIA, the Computational Structural Biology group at the University of Basel & SIB Swiss Institute of Bioinformatics - co-organizers of CASP, and MIT. We aim to provide a gold standard dataset and evaluations to push the field of computational protein-ligand interactions prediction forward.
Expanding the universe of accessible protein-ligand interactions
PLINDER Provides
10x More Data
Paired Unbound & Predicted Structures
High Quality Evaluations
Real-Life Relevant Metrics
Explore PLINDER
Explore Interactively
- Each node is a plinder system with the graph laid out based on shared protein-ligand interactions
- Change the node size and color using different system annotations in the General tab
- Search for properties of interest (domains, oligomeric states etc.) in the Info tab
- Check out subsets of plinder fulfilling numeric constraints in the Analysis tab
- Restrict the date of PDB release using the timeline at the bottom
How PLINDER Was Created
Ingest
- >400k PLI systems across >11k SCOP domains and >50k unique small molecules
- Paired unbound (apo) and AlphaFold2 predicted structures
- 500+ annotations, including protein and ligand properties, quality, and more
- Automated curation pipeline with regular updates
Compare
- Extensive similarity computation across 14 metrics with >20B scores
- Including interaction fingerprints, pocket structural similarity and many more
Split
- Pre-set train/val/test splits that measure generalizability
- Minimizes leakage and maximizes quality of test
- Test subsets with novel proteins, pockets, interactions, ligands
Evaluate
- Retrained existing SOTA methods for comparison
- Standardized eval harness with CASP-CAPRI compatible metrics
- Evaluation on holo, apo and predicted structures
- Leaderboard and Challenges (soon)
Upcoming
- Prediction challenge: PLINDER to be used for PLI challenge at the 2024 NeurIPS ML in Structural Biology (MLSB) workshop
- More data: Measured and predicted binding affinities
- More data: Cryptic pocket and promiscuous ligand annotations
- More data: Data augmentation strategies, including Van-Der-Mers & Cross-Docking, minimized Holo structures to expand Apo coverage
- Leaderboard: Across different tasks and use-cases with multiple upcoming state-of-the-art works already adopted PLINDER
- Regular updates: Enabled through extensive metrics, data ingestion is fully automated and will be updated at regular intervals