Comparative Analysis of Machine Learning Algorithms for Mycobacterium Tuberculosis Protein Sequences on the Basis of Physicochemical Parameters

摘要

The genus Mycobacterium is best known for its two major pathogenic species, M. tuberculosis and M. leprae, the causative agents of two of the world's oldest diseases, tuberculosis and leprosy, respectively. M. tuberculosis kills approximately two million people each year and is thought to latently infect one-third of the world's population. Proteins of these two strains vary a lot and do share some similar characteristics too. Wet lab experiments usually used to classify the proteins of these strains are highly expensive, labor intensive and time consuming. Thus there arises a need for computational approach for classification of H37Rv and Leprae. These computational approaches are fast and economical as compared to wet lab techniques. Realizing their importance, in this paper an attempt has been made to correlate strains with their physicochemical parameters and predict them with fair accuracy. The new SVM learning algorithm called Sequential Minimal Optimization (or SMO) and LibSVM and Multilayer perceptron has been used for classification of H37Rv and Leprae. The performance of the method was evaluated using 5-fold and 10-fold cross-validation. The model has been tested on available data using different SVM learning algorithms and a comparative analysis was made which gives results with 77%, 75% and 85% accuracy respectively. Hence the best algorithm for classification is Multilayer Perceptron using 5-fold cross-validation.