Abstract LB246: Deep learning framework for patient recurrence and genetic marker prediction in diffuse glioma

Abstract

Abstract Investigation of important molecular features of glioma patients may reveal driving mechanisms of patient recurrence and treatment failures in glioma. Applying various deep learning techniques enables the integration and systematic comparison of multi-omics features, allowing robust marker prediction and patient stratification. We extracted mutation and copy number variation data from the Glioma Longitudinal Analysis (GLASS) Consortium, consisting of 628 primary or recurrent samples. To achieve the patient recurrence classification task, we designed a transformer-based model that learns the genomic profiles and predicts patient recurrence status. Our model outperforms traditional machine learning models, including support vector machine (SVM), logistic regression, and decision trees, for up to 34% and 40% improvement of AUROC and AUPRC, respectively. Including mutational features alone can achieve the best prediction performance, while copy number variations are less informative in predicting patient recurrence status. Our approach is an early attempt that employs deep learning for dissecting the contribution of diverse genomic profiles on glioma recurrence and might provide insights into developing novel therapies for treating glioma. Citation Format: Ko-Hong Lin, Kai Zhang, Dung-Fang Lee, Xiaoqian Jiang. Deep learning framework for patient recurrence and genetic marker prediction in diffuse glioma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(7_Suppl):Abstract nr LB246.