TargetNet

Include models with:

Upload SMILES or SDF File

Netting targets for one or more compounds stored in a SMILES or SDF file.

Browse...

Include models with:

Note: Each line of the .smi file should only contain the SMILES string and not contain the compound name.

Target Netting

Save the result as an Excel (.xlsx) file or a tab-separated values (.tsv) file:

Download Excel XLSX Download TSV

Lipinski's Rule of Five

Save the Rule of Five table as an Excel (.xlsx) file or a tab-separated values (.tsv) file:

Download Excel XLSX Download TSV

Target Netting Heatmap

Drug-Target Interation Network Visualization

Network

Model Performance Overview

Tips:

1. Click on the legend labels in the upper right to hide or show the corresponding lines.

2. Move cursor in the chart to see the detailed performance values of each target.

All Targets

A Tutorial on Using TargetNet for Reverse Target Searching

TargetNet User Guide (HTML)

Last revision: November, 2016.

A Brief Introduction of the TargetNet Modeling Strategy

TargetNet is an open-source web application for predicting the binding probability of 623 potential drug targets for given molecule(s).

In drug discovery, one of the big challenges is to identify the potential drug targets for drug-like compounds. However, this could be a difficult task for medicinal chemists. To address such difficulty, we used BindingDB to construct the training sets. BindingDB is a public database of experimentally measured binding affinities, mainly focusing on the interactions of proteins considered to be candidate target with ligands that are small, drug-like molecules.

Activity data were filtered with the following process:

Keep only activity end-point points that had half-maximum inhibitory concentration (IC50), half-maximum effective concentration (EC50) or Ki values;
A compound is considered active when the mean activity value is below 10 uM. All compounds with activity higher than 10 uM are considered inactive;
To ensure that enough number of molecules could be used in model building, only the targets with larger than 200 biological activity data are included.

After this filtering, 109,061 compounds associated with 623 target proteins remained with 115,257 activity end-points, are used for modeling. A set of Random Forest classification models is built using the training set. FP2 fingerprints were computed from the drug-like molecules as features. For each predictive model, a repeated (10 times) 5-fold cross-validation was applied to evaluate the prediction performance. Model performance evaluation metrics include AUC, Accuracy, BEDROC (Boltzmann-Enhanced Discrimination of ROC), MCC, and F-score. Eventually, a model was built with the complete dataset and scored against itself — the training set and whole set should provide similar validation statistics.