Variable selection with a support vector machine for discriminating Cryptococcus fungal species based on ATR-FTIR spectroscopy
Variable selection with supervised classification is currently an important tool for discriminating biological samples. In this paper, 15 supervised classification algorithms based on a support vector machine (SVM) were applied to discriminate Cryptococcus neoformans and Cryptococcus gattii fungal species using ATR-FTIR spectroscopy. These two fungal species of the Cryptococcus genus are the etiological agents of Cryptococcosis, which is an opportunistic or primary fungal infection with global distribution. This disease is potentially fatal, especially for immunocompromised patients, like those suffering from AIDS. The multivariate classification algorithms tested were based on principal component analysis (PCA), successive projections algorithm (SPA) and genetic algorithm (GA) as data reduction and variable selection methods, being coupled to a SVM with different kernel functions (linear, quadratic, 3rd order polynomial, radial basis function, and multilayer perceptron). Some of these algorithms achieved very successful classification rates for discriminating fungal species, with accuracy, sensitivity, and specificity equal to 100% using both SPA-SVM-polynomial and GA-SVM-polynomial algorithms. These results show the potential of such techniques coupled to ATR-FTIR spectroscopy as a rapid and non-destructive tool for classifying these fungal species.