MGMTai: MGMT Methylation Prediction in IDH-Wild Type Glioblastoma (GBM) to Direct Temozolomide (TMZ) Therapy

Abstract Background MGMT promoter methylation status has been utilized as a predictor of response to temozolomide in patients with IDH-wildtype glioblastoma. Traditional methods of methylation status identification include methylation-specific polymerase chain reaction and pyrosequencing. Though widely used, each method has disadvantages with respect to determining methylation cut-off values, tumor content required for evaluation, prognostic accuracy, and financial expense. Methods We have developed a method of MGMT classification using artificial intelligence and a large clinicogenomic database of 5,841 GBM patients. We evaluate the performance of this novel classification strategy for predicting temozolomide treatment efficacy in comparison to pyrosequencing techniques. Result MGMTai reliably predicted MGMT methylation status in GBM patients with available pyrosequencing data. Comparative bucketing of methylation status based on MGMTai and PyroSeq yielded high sensitivity and PPV concordance, though new MGMTmet status percentages were drawn according to PyroSeq methylation percent values. Overall survival with TMZ treatment was comparable between PyroSeq and MGMTai, however, MGMTai by decile and MGMTai stratified into 3 scoring buckets yielded more distinct and predictive survival patterns with increasing MGMTai score compared to PyroSeq. Conclusion Implementation of an AI-based molecular classification system, MGMTai, can better describe MGMT methylation status in comparison to the traditional PyroSeq method. Patient MGMTmet status by MGMTai scoring drew more distinct survival curves and predicted TMZ-treated GBM patient survival more efficiently, less expensively, and with better precision and reproducibility than PyroSeq.