Surface-Enhanced Raman Spectroscopy Of Serum Proteins From Different Digestive System Cancers

Human health is faced with a huge thread by the high incidence and mortality of digestive system cancers. In urgent need of the once screening project that would discriminate various cancers simultaneously. The aim of this study is to develop a more robust dignostic model combining various multivariate statistical analysis methods with surface-enhanced Raman spectroscopy (SERS).The method combined principal component analysis and linear discriminant analysis (PCA-LDA) is used to constructing dignostic model. In this paper, supervised and nolinear dimensionality reduction technologies areintroduced:partial least squares (PLS), kernel principal component analysis (KPCA), Isomap, locally linear embedding (LLE), laplacian eigenmaps (LEM), diffusion maps (DM), multilayer autoencoders (Auc). Then, K-nearest neighbor classifiers (KNN), random forest (RF), artificial neural networks (ANN) and support vector machine (SVM) are intruduced to constructing dignostic model with the low dimentionality representation of SERS. Moreover, we propose an advanced dimensionality reduction method based on the total peak areas (TPA).There are 593 cases of SERS spectra abtained by performing forward and revers ME-SERS Blood plasma samples from three types of digestive system cancers:colorectal cancer (n= 109), gastric cancer (n= 133) and liver cancer (n= 138); and two types of control groups:healthy subjects (n= 85) and nasopharyngeal cancer (n= 128) are analyzed.The major results are as follows:(1) PLS and LLE do better in mapping SERS into their low dimentionality representation with KNN algorithm (10 fold cross-validation). They can control the error rate at 15%,5 percent points lower than PCA method and 7 percent points lower than non-reduction.(2) PLS-SVM does the best when constructing dignostic model, yielding diagnostic accurary of 92.5%, sensitivities of 85.7%,96.4%,92.8%,91.4%, and 94.2%, specificities of 97.6%,95.8%,99.1%,99.1%, and 99.1%, and AUC of 0.91,0.96,0.97,0.96, and 0.97 in the different groups of normal, nasopharyngeal cancers, colorectal cancers, gastric cancers, and liver cancers respectively.(3) TPA is an advanced dimensionality reduction method which yields low dimentionality representation of SERS with physical significance. TPA-SVM does the best when constructing dignostic model, yielding diagnostic accurary of 89.8%; sensitivities of 88.9%,87.5%,81.3%,100%,90.6%; specificities of 96.1%,96.5%, 99.1%,100% and 95.7%; and AUC of 0.89,0.95,0.94,1 and 0.91 in the different groups of normal, nasopharyngeal cancers, colorectal cancers, gastric cancers, and liver cancers respectively.(4) Through experiments, we come to a conclusion that PLS-SVM and TPA-SVM are more robust in constructing dignostic model with SERS spectra. And it was demonstrated serum proteins SERS technology has the potential as a regular physical examination project that can screen various cancers simultaneously only with one blood test.