Abstract 2107: Seeing glycolysis on PDAC: Applying deep learning convolutional neural network model

Abstract

Abstract The reprogramming of cellular metabolism to support continuous proliferation is a hallmark of cancer. Accumulating evidence indicates that the reprogramming of tumor metabolism is under the control of various oncogenic signals. The Kras oncogene is known to induce aerobic glycolysis, and pancreatic adenocarcinoma (PDAC) cells have been shown previously to have metabolism consistent with elevated aerobic glycolysis. Several glycolytic inhibitors are currently in preclinical and clinical development. It had been reported that cancer metabolic status could play a role in the interaction with tumor microenvironment. We hypothesized that cancer glycolysis activity could affect tumor-stromal interaction in PDAC. As a proof-of-concept study, we trained a deep learning convolutional neural network (CNN) model (Google Inception v3) on histopathologic images obtained from The Cancer Genome Atlas (TCGA) PDAC cohort to classify tumors with high and low glycolysis pathway activity. We aimed to answer whether CNNs can predict PDAC glycolysis status using images as the only input. The accuracy on independent validation set is 77.4% with area under ROC curve as 0.698 (Pearson correlation=0.418, p=0.026). Survival analysis showed that higher glycolysis activity was associated with worse clinical outcome in TCGA PDAC. The current study demonstrates the potential of deep learning approaches for histopathologically classifying cancer based on metabolic status. This information could be of value in assisting clinical decisions on targeting cancer metabolism. Citation Format: Peiling Tsou, Chang-Jiun Wu. Seeing glycolysis on PDAC: Applying deep learning convolutional neural network model [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2107.