Predict Protein-Ligand Binding Affinity with GEMS on Vecura

What is GEMS?

GEMS (GNN for Efficient Molecular Scoring) is an advanced graph neural network designed to predict protein-ligand binding affinity directly from 3D structural data. By combining geometric interaction graphs with sophisticated embeddings from pre-trained protein language models (Ankh-Base and ESM2-T6) and a small-molecule language model (ChemBERTa-77M), it offers high-precision affinity scoring. It helps researchers move beyond simple docking, providing a robust, data-driven estimate to triage candidates in drug discovery and virtual screening workflows. It is especially useful for ranking compound libraries when a high-quality protein-ligand pose is already available.

What can users do with GEMS on Vecura?

With GEMS on Vecura, users can:

• Upload 3D Complexes: Easily input PDB-format protein structures and SDF-format ligand poses to initiate scoring.
• Automated Inference: Leverage pre-trained ensemble models without needing to manage complex GPU infrastructure or deep learning environments.
• Customizable Embeddings: Tailor the scoring approach by toggling between different protein language model embeddings (Ankh-Base, ESM2-T6) to optimize performance for specific targets.
• Rank Candidate Compounds: Efficiently generate and interpret pK values to prioritize potent drug candidates for further experimental validation.

What the output means

The output provides a predicted pK value (a numerical estimate of binding affinity on a 0–16 scale) for each provided ligand pose. Higher values signify stronger potential binding, with a pK of 6 typically indicating a micromolar binder and a pK of 9 indicating a nanomolar binder.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

The ability to accurately estimate binding affinity computationally is a cornerstone of modern rational drug design. By integrating protein and chemical language model embeddings into a geometric graph framework, GEMS provides a significant boost in predictive performance over traditional scoring functions. This allows for more effective filtering of large virtual libraries, focusing resources on molecules with a higher probability of success.

For researchers, this translates to faster, more reliable hit identification. By identifying potential binding issues early in the drug discovery pipeline, teams can minimize the time and cost associated with synthesizing and testing compounds that are unlikely to meet binding affinity requirements.

• Developed by: David Graber (camlab-ethz)
• Source: GitHub Repository
• Reference: Preprint: GEMS - A Generalizable GNN Framework For Protein-Ligand Binding Affinity Prediction