DBPP-Predictor

The codes and data for the DBPP-Predictor.

Requipments

• 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

Standalone software available

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.

Molecular representation

In this study, four representation methods were explored including molecular descriptors, molecular fingerprints, molecular graphs and property profiles. They can be implemented as follows:

Molecular descriptors

python descriptor_calc.py

Molecular fingerprints

python FP_calc.py

Property profiles

python Property_Profiles_csv.py

Model training

GNN models

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

Logistic regression model based on QED

python LR_QED.py

DBPP-Predictor

python DBPP_model.py

Trained model

All the models in this study were available in the ‘models’ folder.

DBPP score

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