Abstract

Nicotinic acetylcholine receptors (nAChR) are interesting therapeutic targets due to their involvement in the development of different types of diseases. nAChR inhibitory peptides are considered promising drugs due to their high selectivity and activity on these receptors. However, the identification of nAChR inhibitory peptides using conventional in vitro and in vivo assays is time-consuming and expensive. In this sense, machine learning techniques could offer an advantage to deal with these problems. Among machine learning algorithms, the random forest classifier is one of the best performers in classifying peptides with different types of biological activities. Taking into account the aforementioned aspects, in this work we develop a robust bioinformatic tool for the specific prediction of nAChR inhibitory peptides. In this study, three predictive models with good performance measures were generated from the combination of different features selected using the Gini decrease method and the random forest classifier. The best predictive model presented the following performance measures during the fivefold cross-validation on the training data with Accuracy = 0.85, F1-score = 0.87, Precision = 0.85, Specificity = 0.81, Sensitivity = 0.90, Matthew's correlation coefficient = 0.71; and Accuracy = 0.98, F1-score = 0.98, Precision = 0.95, Specificity = 0.95, Sensitivity = 1.0, Matthew's correlation coefficient = 0.95 in the testing phase. From the selection of the best predictive model, a bioinformatics tool with a friendly user interface was built, called nAChR-PEP-PRED, which allows the analysis of thousands of amino acid sequences. We believe that this tool can accelerate the discovery of new nAChR inhibitory peptides to reduce the time and costs of conventional experimental assays. Our web tool, nAChR-PEP-PRE, is available at https://nachr-pep-pred.herokuapp.com/.