Development of multivariate classification models for the diagnosis of dengue virus infection

The dengue virus (DENV) infection is a worldwide cause of serious illness and death. Early and efficient prediction of disease may help in proper medical management to control disease. Keeping this in view, multivariate classification models by combining with Raman spectroscopy have been developed for the diagnosis of DENV infection in human blood sera. For study design, a statistical analysis is performed to select the sample size for training of models. Total 1240 Raman spectra have been acquired from 39 DENV infected and 23 healthy sera samples. Prior to model development, Raman spectra were examined using ANOVA test for significant differences present in the intensities of newly appeared Raman bands at 622, 645, 700, 746, 800, 814, 873, 890, 948, 1002, 1018, 1080, 1235, 1250, 1272, 1386, 1404, 1446, 1609 and 1645 cm^-1. The significant differences and characteristic patterns of Raman bands induced by disease played decisive role and are exploited for development of multivariate model. Classification models are developed by utilizing principal component analysis (PCA) to extract discriminant features from multidimensional Raman spectral dataset and followed by support vector machines (SVM) with Polynomial of 5, RBF, and liner kernels. The proposed model for this study is built using 10-fold cross validation technique and evaluated on independent dataset to demonstrate its robustness. PCA-SVM (poly-5) model successfully yielded high diagnostic accuracy of 99.52%, sensitivity of 99.75%, specificity of 99.09% for classification of unknown suspected samples. For comparison, PCA discriminant analysis (PCA-DA), partial least squares regression (PLSR) are PLS-DA have been compared. It is found that PCA-SVM (poly-5) approach is more effective and robust compared to other state-of-the-art approaches and it can be used for clinical prediction of DENV infection in human blood sera.