Abstract 4255: TET1 survival model molecular mediation experiments in glioma

Abstract

Abstract Introduction A computational experiment to quantify the molecular mechanisms mediating an association between gene expression and patient survival is described and implemented on TET1 in glioma patients. Methods TET1 has a strong relationship with patient survival in all glioma types. An approach that chains neural networks for expression and survival was developed to investigate the molecular mediators of this relationship. Models were built using Genomic Data Commons RNA-seq data and patient survival glioma data. Multitask gene expression networks were built from TET1 associated gene expression inputs and cancer hallmark gene expression outputs. These networks model the expression of cancer hallmark genes as a function of TET1 associated gene expression. TET1 associated genes were defined as those with the greatest differential expression in TET1 deficient relative to TET1 normal glioma cell lines. Cancer hallmark survival networks were built with RNA-seq expression data inputs and survival data outputs. Input genes were selected from cancer hallmark gene sets. These networks use a cox proportional hazards output layer to train on glioma patient survival data. The models estimate patient hazard as a function of cancer hallmark gene expression. TET1 associated gene expression is transformed into predicted cancer hallmark gene expression via the multitask expression networks. This predicted expression data is then fed into the associated cancer hallmark survival network. By chaining two independently trained cancer hallmark expression and survival networks a ‘mediated concordance’ is measured for each cancer hallmark. Results A new method for interrogating the processes mediating a gene expression / patient hazard correlation is proposed and demonstrated on TET1 and glioma. Immune response, DNA repair and cell motility score highly as mediators of the TET1 survival effect. The DNA repair finding is supported by a positive relationship between TET1 expression and mutation counts derived from Mutation Annotation Format files. The cell motility relationship is supported by a glioma cell line experiment showing greater cell mobility in TET1 deficient cell lines. Conclusions TET1 and genes affected by TET1 expression effectively modeled patient survival with concordance values of 83% in cross validation of the Genomic Data Commons GBM and low-grade glioma data sets. Neural network mediation experiments demonstrated DNA repair mechanisms (median mediated concordance of ~70%) and immune response (median mediated concordance of ~75%) as the most strongly associated hallmarks with the observed TET1 survival effect. Cell motility was inversely associated with the TET1 survival effect with a median mediated concordance of ~25%. Citation Format: Thomas Luechtefeld, Nole Lin, Jua Kim, Channing Paller, Joseph Bressler. TET1 survival model molecular mediation experiments in glioma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 4255.