Abstract Objective The aim of this study was to propose an efficient algorithm to predict antimicrobial peptides using artificial intelligence algorithms. 2 Agricultural Biotechnology Journal; Print ISSN: 2228-6705, Electronic ISSN: 2228-6500 Materials and methods In this study, an updated AMP and non-AMP data set including physico-chemical characteristics at the level of amino acids and protein sequence in three animal species and humans was extracted. After data exploration and data pre-processing steps, four methods Supervised learning including Decision Tree, Random Forest, Naive Bayes and SVM on the AMP dataset with 10-fold cross- validation to build models and predict the AMP label using the evaluation criteria of specificity, sensitivity, rate Accuracy, precision, recall, F1 score and area under the rock curve (AUC) were evaluated. Results In this study, using an up-to-date dataset, a machine learning model has been successfully trained to predict antimicrobial peptides. A comprehensive set of features has been subjected to feature selection to identify key features of antimicrobial peptides. After selecting the feature, among the different generated models, the model based on the RF model classifier showed the best performance with Accuracy (95 percent), Precision (96 percent), Recall (95 percent), F1 Score (95 percent). the four classification of algorithms, Random Forest algorithm and SVM are the most accurate. The Decision Tree classification algorithm had the least accuracy. Conclusions According to the obtained results, the proposed RF model has a better performance than other models for AMP prediction. This model predicted some peptides as peptides with antimicrobial properties. This predictive approach can be useful in extracting AMPs with antimicrobial properties from AMP libraries in useful clinical applications before moving on to experimental studies. On the other hand, several features in the final selection properties indicate that these features are critical determinants of peptide properties and should be considered in the development of models to predict peptide activity. The executable code is available in the attached file

کلمات کلیدی