GHDDI's Free Structure based Prediction Service

To systematically assess the structure based model properties:

1. Please paste your SMILES list, Drug Name list or upload a file here. (example) Each query can handle 100 compounds at most.

Select input file(*.csv,*.xls, *.xlsx)

2. Select 1 or more property category listed below.

3. Please upload a pdb file here.

Select PDB file

4. Select the output report type.

Model Documentation

Structure based (none-docking) AI model

The structure-based AI model was constructed based on GHDDI’s in-house developed HAG-net. The model is universal for all targets with 3D structures. The model was evaluated using the DUD.E benchmark set with an average AUC of 0.99; after removal of computer generated decoys, the model showed an average AUC of 0.945. The performance is based on 2 million protein-ligand interactions of over 185 protein targets. Given a target 3D structure, preferably the binding pocket with a radius of 15 Å, and screening library SMILES list as input, we are able to screen 14 million compounds per day on a single GPU, saving over 1000 times of computational cost compared to traditional docking tools. The model output is a probability distribution between 0 and 1 where a predicted value closer to 1 predicts a greater probability of docking.