Application of machine learning techniques towards classification of drug molecules specific to peptide deformylase against Helicobacter pylori

It is crucial to adapt to the current computational drug discovery pipeline to develop novel drug molecules to combat the gastric disorders caused by Helicobacter pylori. Virtual screening techniques can be used as a preliminary screening tool to identify the relevant compounds which may have drug-like properties. These drug-like molecules can be further screened to test their bioactivity against a particular protein target. In this context, we apply different machine learning techniques to generate models to predict the pIC50 value of drug molecules. Molecular descriptors were produced for the drug data set. Initial models were developed for the data set with a large number of descriptors. Later, feature reduction techniques were applied to yield feature descriptors with the best six variables using three algorithms: principal component analysis, random forest and genetic algorithm. Consequently, machine learning techniques were applied to the reduced data set to develop predictive models. Nai ve Bayes algorithm achieved better accuracy of 84.42% compared with other algorithms. The results were validated on the test set using 10-fold cross validation. The methodology can be applied to predict the bioactivity of drug molecules. The procedure can be further implemented to identify novel drug molecules against pathogenic H. pylori by blocking its functionalities. The computational process also helps reduce the timeline of drug discovery process.