The codes and data for the DBPP-Predictor.
- python 3.7
- descriptastorus 2.2.0
- Pytorch 1.5.0+
- dgl-cu101 0.7.0
- dgllife 0.2.8
- rdkit 2021.03.4
- scikit-learn
- numpy
- pandas
The DBPP-predictor standalone software is available at https://www.amazon.com/clouddrive/share/f9d5ZQk6UE5ayGSKfnZJu93Cg2SSQE4el9SMM7aZpUK (Amazon Drive)
or
https://figshare.com/articles/software/DBPP-Predictor_standalone_software/23813805 (Figshare)
Guidance:
1.download: Full version standlone software available at Amazon Drive and Figshare repositories.
2.Access: The DBPP-Predictor standalone software does not need to be installed, users can start using it by double clicking DBPP_Software\dist\DBPP_Predictor.exe file.
3.Usage: Detailed instructions can be obtained through the "Help" button of the software.
In this study, four representation methods were explored including molecular descriptors, molecular fingerprints, molecular graphs and property profiles. They can be implemented as follows:
python descriptor_calc.py
python FP_calc.py
python Property_Profiles_csv.py
The four GNN models can be trained as follow:
python AttentiveFP_classify.py
python GAT_classify.py
python GCN_classify.py
python GraphSAGE_classify.py
python LR_QED.py
python DBPP_model.py
All the models in this study were available in the ‘models’ folder.
The DBPP score can be utilized for the assessment of drug-like properties of new compounds. It can be implemented as follows:
python Model_Validation.py