Abstract Introduction: Glioma is an aggressive brain tumor. Many inflammatory mediators within the tumor microenvironment (TME) determine its progression. Unfortunately, no ubiquitous prognostic markers reveal the intricacy between TME and chronic inflammatory responses (CIRs). CIRs used to vary among individuals affecting prognosis. And opting for the best-optimized treatment remains challenging. Methods: We proposed an explainable artificial intelligence (XAI) approach to redefine CIRs as prognostic markers for glioma using 694 and 693 patients’ data derived from The Cancer Genome Atlas (TCGA) and the Chinese Glioma Genome Atlas (CGGA), respectively. First, using five curated databases, we identified CIR-related genes those enriched nine HALLMARK pathways using gene set variation analysis (GSVA). Then, we trained the XAI model to characterize the personalized inflammatory mediators (PIMs) using multimodal factors such as transcription factors (TFs) network and cell infiltration markers. It resulted in rationalized typified CIRs. Finally, we used overall survival (OS) analyses to validate them as potential prognostic markers and showed a personalized co-drug discovery strategy by optimizing CIRs. Results: The proposed XAI model deciphered the PIMs and endorsed the typified CIRs as efficient prognostic markers. It characterized five unique CIR types (p-value<0.0001). The OS analyses indicated the differential prognostic ability of the typified CIRs for glioma, lower-grade glioma (LGG) at p-value<0.0001, and glioblastoma (GBM) at p-value<0.22. We found that patients with higher CIRs had significantly shorter OS than those with lower CIRs. We showed that CIR type 4 and type 2 were highly inflamed and had poor OS for LGG, and GBM, respectively. Chronic pro-inflammatory responses were positively correlated with the poor OS of glioma patients. GSVA indicated lower activation of apoptotic pathways and higher activation of the hypoxia pathway as likely causes. We identified the effector genes within the personalized CIR networks, which demonstrated prognostic ability and may offer potential targets for optimizing the patient-specific CIRs through co-drug intervention. Additionally, we identified their transcriptional regulatory networks comprising TFs (e.g., SMAD3, HIF1A, NFKB1, GATA3, RORA, JUN) contributing to the inflammatory pathways (e.g., TNFα, NF-kβ, IFN-γ, hypoxia). Conclusions: Our findings highlight a new strategy to redefine CIRs as prognostic markers for glioma and show that optimizing CIRs improves the OS. It opened a new avenue for co-drug discovery to ensure better survival. Acknowledgments: HK GRF (12102722, 12101018, 12102518, 12100719), HK Theme-based Scheme (T12-201/20-R), IRMS-HKBU (RC-IRMS/15-16/01), TISSF (Guangdong-Hong Kong-Macau Joint Lab, 2020B1212030006) and NSFS (ZR2020QH219). Citation Format: Debajyoti Chowdhury, Hiu Fung Yip, Hao Liu, Xuecheng Tai, Aiping Lu. Redefined chronic inflammatory responses as prognostic markers in glioma: an explainable artificial intelligence model. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5412.
Read full abstract