Machine Learning-based Virtual Screening for STAT3 Anticancer Drug Target.

Area of Science:

• Computational chemistry and bioinformatics
• Drug discovery and development
• Molecular biology and oncology

Background:

• Signal transducers and activators of transcription (STAT) proteins, particularly STAT3, are crucial in cancer progression.
• STAT3 hyperactivation is observed in numerous cancers, making it a significant therapeutic target.
• Traditional drug discovery is lengthy and expensive; machine learning offers a faster alternative for identifying drug candidates.

Purpose of the Study:

• To conduct machine learning-based virtual screening for novel STAT3 inhibitors.
• To identify and validate potential drug candidates targeting the STAT3 signaling pathway.

Main Methods:

• Development and validation of machine learning models (k-NN, SVM, Naïve Bayes, Random Forest) for classifying STAT3 inhibitors.
• Virtual screening of a compound library using the Random Forest model.
• Molecular docking and 100ns molecular dynamics (MD) simulations for top-ranked compounds.

Main Results:

• The Random Forest model demonstrated superior performance in classifying active and inactive STAT3 inhibitors.
• Virtual screening identified 20 compounds with 88% accuracy as potential STAT3 inhibitors.
• Top predicted compounds exhibited greater stability and compactness compared to the reference compound in MD simulations.

Conclusions:

• The identified compounds show promise as STAT3 inhibitors for treating STAT3-associated diseases.
• Machine learning-driven virtual screening is an effective strategy for accelerating the discovery of novel cancer therapeutics.
• These findings contribute to the development of targeted therapies for various malignancies.