Prediction of Cancer Metastasis Using SVM-Based Model with Quantification of Pathway Changes between Different Disease Statuses

Discriminating different subtype is critical for cancer treatment strategy. In-silico approaches are usually confined to use gene expression profile. In previous studies for this topic, researches usually focused on the gene level or specific disease, thus factors in the pathway level are not considered. We use a scored equation that integrates genomics and proteomics information. The quantification of the strength of pathway link change is implemented. A support vector machine (SVM) is used to train and test subtype-predicted models. The average prediction accuracy reaches 67.641058% for three items in tumors of neuroepithelial tissue. We have devised a new method of subtype prediction with novel perspective, and demonstrated that this method can apply features far less than only gene-expression used to obtain similar result. This study suggests a way for implementing a cancer subtype classifier based on SVM from pathway-level view.