A Multi-Model Machine Learning Framework for Identifying Raloxifene as a Novel RNA Polymerase Inhibitor from...

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning in pharmacology

Background:

• RNA-dependent RNA polymerase (RdRP) is a key target for antiviral drug development.
• Identifying novel RdRP inhibitors is crucial for combating viral infections.

Purpose of the Study:

• To develop and validate a machine learning framework for identifying RdRP inhibitors.
• To screen FDA-approved drugs for potential RdRP inhibitory activity.
• To evaluate raloxifene as a potential RdRP inhibitor using computational methods.

Main Methods:

• Ensemble machine learning models (ExtraTreesClassifier, RandomForestClassifier, LGBMClassifier, BernoulliNB, BaggingClassifier) combined with a Convolutional Neural Network (CNN).
• Utilized PubChem dataset AID 588519 for training and validation.
• Performed molecular docking and molecular dynamics simulations on norovirus RdRP (PDB: 4NRT).

Main Results:

• Ensemble models achieved accuracy, ROC-AUC, and F1 scores >0.70.
• CNN model showed a specificity of 0.77 on external validation.
• Raloxifene exhibited a favorable binding affinity (-8.8 kcal/mol) and stable binding dynamics with norovirus RdRP.

Conclusions:

• The developed machine learning framework effectively identifies potential RdRP inhibitors.
• Raloxifene is a promising candidate for RdRP inhibition and warrants experimental validation.
• Findings support raloxifene's potential as a broad-spectrum antiviral agent.