A 80-gene set potentially predicts the relapse in laryngeal carcinoma optimized by support vector machine.

Abstract

The present study was performed to identify a gene set for predicting the relapse in laryngeal carcinoma using large data analysis methods. Two gene expression profile data of laryngeal carcinoma (GSE27020 and GSE25727) were downloaded from public database. Genes associated with tumor relapse, namely informative genes, were identified by Cox regression analysis. Then the protein-protein interaction (PPI) network consisting of informative genes was constructed. Afterwards, the optimized support vector machine (SVM) classifier was constructed to classify the relapsed laryngeal carcinoma samples based on genes in specific PPI network. Furthermore, the efficiency of the SVM classifier was verified by other two independent datasets. A total of 331 informative genes were obtained from GSE27020 and GSE25757 datasets. A PPI network specific to laryngeal carcinoma relapse was constructed which contained informative genes and critical non-informative genes. The top 10 genes in specific PPI network were APP, NTRK1, TP53, PTEN, FN1, ELAVL1, HSP90AA1, XPO1, LDHA and CDK2 ranked by BC (betweenness centrality) value. The optimized SVM classifier including top 80 genes showed accuracy of 100% to classify the relapsed cases from laryngeal carcinoma samples. Next, the efficiency of the SVM classifier to predict relapse samples was verified in another independent datasets, which showed accuracy of 97.47%. The informative genes in the optimized SVM classifier were enriched in several pathways associated with tumor progression. A 80-gene set was identified as biomarker to predict the relapse of laryngeal carcinoma, which would be potentially applied in decision of different treatments for patients with different relapse risks.