Disease-Ligand Identification Based on Flexible Neural Tree.

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Network pharmacology research requires accurate screening of disease-related compounds in traditional Chinese medicine (TCM).
• Existing virtual screening methods may lack the necessary precision for complex TCM prescriptions.

Purpose of the Study:

• To develop a novel virtual screening method for accurately identifying disease-related compounds in TCM.
• To enhance the efficiency and reliability of network pharmacology research.

Main Methods:

• Proposed a new virtual screening method integrating a flexible neural tree (FNT) model, a hybrid evolutionary algorithm (combining Grammar-guided genetic programming and salp swarm algorithm), and a negative sample selection algorithm.
• Collected disease-related compounds for hypertension, diabetes, and COVID-19 from literature.
• Utilized ECFP6, MACCS, Macrocycle, and RDKit for chemical structure characterization.

Main Results:

• The proposed FNT-based method demonstrated superior performance compared to classical classifiers (SVM, RF, AdaBoost, DT, GBDT, KNN, LR, NB), gcForest, and forgeNet.
• Performance was evaluated using metrics including AUC, ROC, TPR, FPR, Precision, Specificity, and F1.
• The MACCS molecular descriptor proved suitable for the maximum number of classifiers, while ECFP6 showed poor performance across all methods.

Conclusions:

• The developed hybrid evolutionary FNT model offers a more accurate and effective approach for virtual screening of TCM compounds.
• This method significantly advances network pharmacology by improving the identification of relevant therapeutic agents.
• The findings highlight the importance of appropriate molecular descriptor selection for successful virtual screening.