01 / UniMolRep
Explore a molecule
- Start with
- A molecule as SMILES
- Receive
- Fingerprints, atom graphs, geometry and spatial grids
Understand the structure a model sees.
Explore representationsFrom molecular structure to binding affinity
Explore molecular representations, screen against a protein target and score a bound pose using your own inputs. Discover how ETDock predicts binding poses, and connect each method to the research behind it.
Choose your taskFour methods, from molecules to complexes
02 / DL-FSG
Screen for a target
- Start with
- SMILES and a protein target
- Receive
- A relative screening score with Grad-CAM attribution.
Compare molecules within the same target.
Screen a molecule03 / ETDock
Predict a binding pose
- Start with
- A protein structure and a ligand
- Receive
- A predicted three-dimensional ligand pose
Predict ligand poses in your browser.
Explore docking
04 / GNINA / BAP
Score a bound pose
- Start with
- A PDB ID or your own protein–ligand complex
- Receive
- A GNINA affinity estimate and pose score with spatial occlusion.
Choose the protein chains and bound ligand.
Score a complexFrom input to insight
- 01Submit your inputChoose a task and enter your molecule or complex.
- 02Explore the resultUse the molecular views and concise score explanations.
- 03Take it with youDownload data, structures and a readable run summary.
- 04Go deeperFollow each method to the research behind it.
Research on molecular structure and binding
Our research connects molecular representation with three questions: which compounds to prioritise, how they bind, and how strongly they may bind.
UniMolRep2026
Represent molecular structure
Fingerprints, graphs and geometry capture different information. UniMolRep studies how these descriptions support molecular learning.
DL-FSG2026
Prioritise candidate compounds
Learning from sequence and spatial features together helps rank molecules for a protein target. DL-FSG studies this ligand-based screening task.
ETDockOnline 2026 (2027 issue)
Predict a binding pose
A candidate molecule’s position and orientation determine how it meets a protein. ETDock learns distances between the partners to reconstruct a binding pose.
AGIMA-Score2025
Learn from the binding interface
Given a protein–ligand complex, AGIMA-Score studies how contacts between the partners can inform an estimate of binding strength.
Research project
Protein-ligand Binding Affinity Prediction Based on Grid Representation and Deep Learning: Paving the Way to Efficient Structure-based Drug Design
The project investigates how molecular grids and deep learning can improve the accuracy, interpretability and efficiency of protein–ligand affinity prediction. Its aim is to support structure-based drug design through better virtual screening and rescoring of protein–ligand poses.
Calculations run in your browser. Molecules and uploaded complexes stay on your device; PDB lookups request the selected structure from the Protein Data Bank.