Analysis of differentially expressed M. fortuitum proteins for biomarker prediction using Support Vector Machine

Non-tuberculous Mycobacterium fortuitum is a bacterium that causes opportunistic infections in humans. Skin, joint, traumatic, pulmonary, and biofilm-related infections are among the principal clinical manifestations of the pathogen. Drug-resistant infections caused by M. fortuitum necessitate prolonged treatment including several medications. Our knowledge of its reservoirs, modes of transmission, virulence factors, and conventional treatment protocols is lacking. The gap in our understanding of the molecular and phenotypic evolution justifies the need to discover and investigate proteins differentially expressed in the bacteria under different growth and multiplication states. Global proteome analysis of the bacteria was done in triplicate under planktonic state. Three planktonic datasets were subjected to Support Vector Machine, a machine learning technique for biomarker prediction. The performance parameters were assessed for the models and the optimal values for the cost function, gamma, and CV were evaluated. The model was validated using kNN and found that SVM shows better prediction results. For the P1 dataset, there was an accuracy improvement of 7.44% over SVM in comparison to kNN; for the P2 and P3 datasets, the improvements were 8.80% and 6.39%, respectively. The study and analysis are a novel attempt to analyze protein expression data obtained for a human pathogen which may lead to prediction of diagnostic and therapeutic biomarkers.