Prediction of Mycobacterium tuberculosis cell wall permeability using machine learning methods.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning in medicine

Background:

• Tuberculosis (TB) remains a global health challenge, exacerbated by drug-resistant strains.
• The unique Mycobacterium tuberculosis (M. tb) cell wall presents a significant barrier to drug penetration.
• Novel therapeutic strategies require compounds capable of traversing this protective layer.

Purpose of the Study:

• To develop and validate a reliable machine learning model for predicting mycobacterial cell wall permeability.
• To identify key molecular descriptors influencing permeability for guiding future drug design.

Main Methods:

• Trained four machine learning algorithms (Random Forest, SVM, k-NN, Logistic Regression) on a dataset of 5368 compounds.
• Utilized RDKit and Mordred toolkits for molecular feature calculation.
• Evaluated models using accuracy, precision, recall, F1 score, and AUC; refined the best model with hyperparameter tuning and cross-validation.

Main Results:

• The Support Vector Machines (SVM) model with filtering demonstrated superior performance.
• Achieved 80.26% accuracy on the test set and 81.13% on the validation set.
• Identified critical molecular descriptors that govern M. tb cell wall penetration.

Conclusions:

• The developed SVM model effectively predicts mycobacterial cell wall permeability.
• This predictive model can accelerate the design and discovery of new anti-TB drugs.
• The insights into permeability-driving features offer valuable guidance for medicinal chemists.