YIP, H. F. | bioRxiv Science

bioRxiv2023

An explainable artificial intelligence-based typification of chronic inflammatory responses enhances glioma prognosis

Glioma is one of the most aggressive solid brain tumors with a poor prognosis. A chronic tumor inflammatory microenvironment drives glioma promotion and progression. The neutrophil-to-lymphocyte ratio and other clinicopathological variables usually serve as prognostic glioma markers. However, they are not ubiquitous prognostic markers for glioma as they fail to reveal the intricacy between the glioma-specific tumor inflammatory microenvironment and the systemic inflammatory responses, especially those chronic inflammatory responses, which vary among individuals fabricating diverse prognostic outcomes. Here, we introduced an explainable artificial intelligence model to typify chronic inflammatory responses as prognostic markers for glioma using 694-patients data from The Cancer Genome Atlas. We characterized the glioma-specific personalized inflammatory mediators using multi-layered regulators such as transcriptional networks, cellular infiltration markers, and cellular senescence markers, which identified five unique chronic inflammatory responses (p-value<0.0001). We defined its prognostic significance using overall survival analyses. The chronic inflammatory responses were positively correlated with poor overall survival in glioma. The patients with higher chronic inflammatory responses showed significantly shorter overall survival than those with lower chronic inflammatory responses. Interestingly, optimizing those chronic inflammatory responses improved the overall survival of glioma patients. We identified the effector genes within the personalized inflammatory mediators networks, indicating them as the targets for optimizing individualized chronic inflammatory response profiles through co-drug intervention. SignificanceExplainable artificial intelligence-based typification of chronic inflammatory responses accelerates glioma prognosis and supports co-drug discovery to modulate inflammatory responses alongside cancer therapy, suggested by 694-glioma patients data analysis.

systems biology↗

bioRxiv2022

ReDisX: a Continuous Max Flow-based framework to redefine the diagnosis of diseases based on identified patterns of genomic signatures

Diseases originate at the molecular-genetic layer, manifest through altered biochemical homeostasis, and develop symptoms later. Hence symptomatic diagnosis is inadequate to explain the underlying molecular-genetic abnormality and individual genomic disparities. The current trends include molecular-genetic information relying on algorithms to recognize the disease subtypes through gene expressions. Despite their disposition toward disease-specific heterogeneity and cross-disease homogeneity, a gap still exists to describe the extent of homogeneity within the heterogeneous subpopulation of different diseases. They are limited to obtaining the holistic sense of the whole genome-based diagnosis resulting in inaccurate diagnosis and subsequent management. To fill those gaps, we proposed ReDisX framework, a scalable machine learning algorithm that uniquely classifies patients based on their genomic signatures. It was deployed to re-categorizes the patients with rheumatoid arthritis and coronary artery disease. It reveals heterogeneous subpopulations within a disease and homogenous subpopulations across different diseases. Besides, it identifies GZMB as a subpopulation-differentiation marker that plausibly serves as a prominent indicator for GZMB-targeted drug repurposing. The ReDisX framework offers a novel strategy to redefine disease diagnosis through characterizing personalized genomic signatures. It may rejuvenate the landscape of precision and personalized diagnosis, and a clue to drug repurposing.

bioinformatics↗