Fig. 5: Predicting brain tumor subtypes using sscf-MeDIP-Seq datasets.
From: Tumor detection by analysis of both symmetric- and hemi-methylation of plasma cell-free DNA
a A workflow for building brain tumor subtype models. Models for the IDH WT and IDH mutant gliomas were first trained by DMRs and DHMRs identified using the training cohort samples and then combined with the three class models (controls, liver and brain tumor) based on the Bayes’ theorem to derive models for predicting four sample groups: IDH WT and IDH mutant brain tumor, liver tumor and control samples. Evaluation of predicting IDH mutant (b) and IDH wild type (c) brain cancer samples in validation cohort using models trained with DMRs, DHMRs, DMRs+DHMRs. The best sensitivity and specificity point are labeled as red dots on the curve. The 95% confidence interval of AUC for each model is labeled in parenthesis. d The average prediction probability of each group of samples based on DMR+DHMR models. Each column represents the sample groups in the validation cohort, with each row representing model predictions. Bar plots are presented as mean value + standard error. Red, pink, yellow and blue bars represent probability of samples from 11 IDH mutant brain cancer, 9 IDH WT brain cancer, 15 liver cancer, and 21 controls, respectively.
