DL-DILI Prediction Server

Introduction of DL-DILI
Drug-induced Liver Injury (DILI) is the single most frequent cause of safety-related drug marketing withdrawals. DL-DILI webserver uses deep learning (DL) methods to predict DILI. Two DL-DILI models (DL-Combined and DL-Liew), which were developed based on different datasets and DILI annotations are available in this server. The DL-combined model trained on 451 (original: 475) drugs and predicted an external validation set of 198 drugs with an accuracy of 86.4% (original: 86.9%), sensitivity of 83.3% (original: 82.5%), specificity of 90.5% (original: 92.9%), and area under the curve of 0.935 (original: 0.955), which is better than the performance of previously described DILI prediction models. The DL-Liew model trained on 1065 compounds with higher diversity than DL-combined and tested 119 compounds with an accuracy of 77.3%, sensitivity of 81.4%, specificity of 71.4% and area under the curve of 0.855. The revised dataset can be downloaded here.

Input molecule data
Input File
Model Type

1. The format of an input file is allowed, such as "*.smi", "*.mol", "*.mol2", "*.sdf"; The size of an input file is less than 200KB;
2. The"input File" option allows you to upload a file including one or more molecules (the number of heavy atoms must be less than 100).
3. It is recommended to use OpenBabel toolkit (http://openbabel.org/wiki/Main_Page) to format the SMILES files for better and faster prediction.
4. DL-Liew model: Predicting DILI of compounds using Liew et al.'s DILI annotation (Liew et al., Journal of Computer-Aided Molecular Design 2011, 25, 855-871).
5. DL-Combined model: Predicting DILI of pharmaceutical compounds using Chen et al.'s DILI annotation (Chen et al., Toxicological Sciences 2013, 136, 242-249).
6. This web server is freely accessible to academics. For industry, please send inquires to jfpei@pku.edu.cn.