QSAR Models for Active Substances against Pseudomonas aeruginosa Using Disk-Diffusion Test Data.

Area of Science:

• Microbiology
• Medicinal Chemistry
• Computational Chemistry

Background:

• Pseudomonas aeruginosa is a critical Gram-negative pathogen and one of the ESKAPE group of microbes, known for antibiotic resistance.
• Developing novel antibiotics against P. aeruginosa is a high priority due to its significant threat to public health.
• Quantitative structure-activity relationship (QSAR) models typically use minimum inhibitory concentration (MIC) data, but disk diffusion data has not been extensively explored for this purpose.

Purpose of the Study:

• To investigate the utility of disk diffusion results (inhibition zones) for developing QSAR models against Pseudomonas aeruginosa.
• To compare the performance of various machine learning algorithms in predicting antibiotic activity using disk diffusion data.

Main Methods:

• Employed multiple machine learning algorithms including support vector classifier, K nearest neighbors, random forest, decision tree, AdaBoost, logistic regression, and naive Bayes.
• Utilized four distinct sets of molecular descriptors and fingerprints, alongside three data balancing techniques and the original dataset.
• Built 32 individual models for each descriptor/fingerprint and balancing method combination, then stacked 28 of these into meta-models.

Main Results:

• K nearest neighbors (KNN), logistic regression, and decision tree classifiers demonstrated strong performance in terms of balanced accuracy.
• Ensemble methods, created by stacking multiple models, showed slightly superior predictive results when evaluated using nested cross-validation.
• The study successfully demonstrated the feasibility of using disk diffusion data for QSAR model development.

Conclusions:

• Disk diffusion data is a viable alternative to MIC values for developing QSAR models against Pseudomonas aeruginosa.
• Machine learning, particularly ensemble methods, can effectively leverage disk diffusion data for antibiotic discovery efforts.
• This approach offers a promising avenue for accelerating the development of new antibiotics against resistant Gram-negative bacteria like P. aeruginosa.