Predicting Cancer Metastasis using DNA Methylation and Gene Expression Profiles

초록

Metastasis is the major cause of cancer-related mortality, accounting for about 90% of cancer deaths. So far, most computational methods for predicting metastasis used either gene expression data or relation between genes without accounting for epigenetic mutations to gene expression. The association of DNA methylation with gene expression in cancer remains to be fully identified, but recent studies reported that there are inter-person variations in DNA methylation as well as in gene expression. Motivated by increasing evidence of inter-person variations not only in gene expression but also in DNA methylation, we developed a new method for predicting metastasis from gene expression and DNA methylation profiles.?In this study, we computed changes in correlations between DNA methylation and gene expression and in each tumor sample of several cancer types. We constructed logistic regression models with different combinations of features for predicting lymph node metastasis and distant metastasis. Testing the models on independent datasets which were not used in training showed that differential correlations of DNA methylation with gene expression in individual patients or DNA methylation beta values are more powerful features than gene expressions in predicting metastasis. The features identified in our study and our approach will be useful for predicting metastasis, which in turn will help select treatment options for cancer patients.