Prediction of Novel Antibiofilm Peptides from Diverse Habitats using Machine Learning

Abstract

Multidrug resistant bacteria often lead to biofilm formation. Biofilm is a colonizedform of pathogens (fungi, bacteria) attached to surfaces like animal or plant tissues, medical devices like catheters, and artificial heart valves. Biofilm formation prolongs the survival of microorganisms in an adaptive environment, leading to the spread of infection in different organs and causing a high morbidity rate. Given the rise of chronic infection and antibiotic resistance due to biofilm, it is essential to find an alternative solution to control biofilm infections. Antibiofilm peptides can interact with these biofilm-creating pathogens to inhibit growth, virulence, and biofilm formation. We hypothesized that mining the existing peptide databases from diverse habitats could provide potential antibiofilm activities for our work. We developed a computational model to predict the antibiofilm properties by applying machine learning algorithms like support vector machine, random forest, extreme gradient boosting, and multilayer perceptron classifier. We evaluated more than 240 antibiofilm peptides and more than 570 different compositions and motif-based features to build our prediction model. We also created a regression model on top of our classifier to predict the effectiveness of peptides by curating minimum inhibitory concentration against biofilm. Our classifiers achieved greater than 98% accuracy while the harmonic mean of precision-recall (F1) and Matthews correlation coefficient (MCC) scores obtained are greater than 0.91. Using this two-tier model approach, we assessed more extensive databases of antimicrobial, anticancer, antiviral, and dairy peptides for potential antibiofilm functionality and came up with the top ten potential candidates of antibiofilm peptides.