Elucidate the mechanism underlying resistance in gliomas to conventional therapy

The objectives of our study are to elucidate the mechanism underlying resistance in gliomas to conventional therapy (e.g. ionizing radiation and temozolomide) and to develop a model to better predict patients with gliomas that have a higher risk of failing conventional therapy.  Based on data from our laboratory, we hypothesize that resistance is due to low levels of expression of Tet Methylcytosine Dioxygenase 1 (Tet1) resulting in poor DNA repair and increased genomic instability. To test the hypothesis, we will use data from the Genomic Data Commons (GDC) application programming interface to extract mutation, copy number, differential expression, and clinical events. Biological features relevant to both survival prediction and Tet1 expression will be selected using genetic annotation tools such as paxtools, biomart and others as necessary. Focusing on molecular features with known mechanistic relationships to TET1 allows the construction of models that are less likely to be overfit and will generate data to identify new targets for drug development through laboratory experiments.

To achieve our second objective, the data extracted from the GDC will be loaded into formats for survival analysis. Traditional statistical methods will be used such as those implemented in the CRAN “survival” package.  Following traditional approaches, we will apply machine learning approaches with a focus on recursive neural networks (implemented in deeplearning4j).  Recursive Neural networks are ‘time aware’ machine learning models that can account for survival analysis intricacies such as latency effects, repeat measurements, and censoring.  The ability to predict patient responses will be useful in developing a personalized approach for treating gliomas.