Abstract PO-046: MethylationToActivity: A deep-learning framework that reveals promoter activity landscapes from DNA methylomes in individual tumors

Although genome-wide DNA methylomes have demonstrated their clinical value as reliable biomarkers for tumor detection, subtyping, and classification, their direct biological impacts at the individual gene level remain elusive. Here we present MethylationToActivity (M2A), a machine learning framework that uses convolutional neural networks to infer promoter activities based on H3K4me3 and H3K27ac enrichment, from DNA methylation patterns for individual genes. Using publicly available datasets in real-world test scenarios, we demonstrate that M2A is highly accurate and robust in revealing promoter activity landscapes in various pediatric and adult cancers, including both solid and hematologic malignant neoplasms. Citation Format: Justin Williams, Beisi Xu, Daniel Putnam, Andrew Thrasher, Chunliang Li, Jun Yang, Xiang Chen. MethylationToActivity: A deep-learning framework that reveals promoter activity landscapes from DNA methylomes in individual tumors [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-046.