CIDER: Counterfactual-Invariant Diffusion-based GNN Explainer for Causal Subgraph Inference DOI Creative Commons

Q. Zhang,

Chengshang Lyu, Lingxi Chen

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 9, 2024

Abstract Inferring causal links or subgraphs corresponding to a specific phenotype label based solely on measured data is an important yet challenging task, which also different from inferring nodes. While Graph Neural Network (GNN) Explainers have shown potential in subgraph identification, existing methods with GNN often offer associative rather than insights. This lack of transparency and explainability hinders our understanding their results underlying mechanisms. To address this issue, we propose novel method link/subgraph inference, called CIDER: Counterfactual-Invariant Diffusion-based ExplaineR, by implementing both counterfactual diffusion implementations. In other words, it model-agnostic task-agnostic framework for generating explanations counterfactual-invariant process, provides not only due implementation but reliable the process. Specifically, CIDER first formulated as inference task that generatively two distributions one another spurious subgraph. Then, enhance reliability, further model Thus, using distribution, can explicitly quantify contribution each phenotype/label manner, representing subgraph’s strength. From causality perspective, interventional method, traditional association studies observational approaches, reduce effects unobserved confounders. We evaluate synthetic real-world datasets, all demonstrate superiority over state-of-the-art methods.

Language: Английский

Prediction model for spinal cord injury in spinal tuberculosis patients using multiple machine learning algorithms: a multicentric study DOI Creative Commons
Sitan Feng, Shujiang Wang, Chong Liu

et al.

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: April 2, 2024

Abstract Spinal cord injury (SCI) is a prevalent and serious complication among patients with spinal tuberculosis (STB) that can lead to motor sensory impairment potentially paraplegia. This research aims identify factors associated SCI in STB develop clinically significant predictive model. Clinical data from at single hospital were collected divided into training validation sets. Univariate analysis was employed screen clinical indicators the set. Multiple machine learning (ML) algorithms utilized establish models. Model performance evaluated compared using receiver operating characteristic (ROC) curves, area under curve (AUC), calibration analysis, decision (DCA), precision-recall (PR) curves. The optimal model determined, prospective cohort two other hospitals served as testing set assess its accuracy. interpretation variable importance ranking conducted DALEX R package. deployed on web by Shiny app. Ten characteristics for random forest (RF) emerged choice based AUC, PRs, DCA, achieving test AUC of 0.816. Additionally, MONO identified primary predictor through ranking. RF provides an efficient swift approach predicting patients.

Language: Английский

Citations

5
A predictive model for early clinical diagnosis of spinal tuberculosis based on conventional laboratory indices: A multicenter real-world study DOI Creative Commons
Xiaojiang Hu, Guang Zhang, Hongqi Zhang

et al.

Frontiers in Cellular and Infection Microbiology, Journal Year: 2023, Volume and Issue: 13

Published: March 24, 2023

Background Early diagnosis of spinal tuberculosis (STB) remains challenging. The aim this study was to develop a predictive model for the early STB based on conventional laboratory indicators. Method clinical data patients with suspected in four hospitals were included, and variables screened by Lasso regression. Eighty-five percent cases dataset randomly selected as training set, other 15% validation set. diagnostic prediction established logistic regression nomogram drawn. performance verified Result A total 206 included study, including 105 101 NSTB. Twelve modeled regression, seven (TB.antibody, IGRAs, RBC, Mono%, RDW, AST, BUN) finally model. AUC 0.9468 0.9188 cohort, respectively. Conclusion In we developed which consisted routine

Language: Английский

Citations

7
IL-10 and TGF-β1 may weaken the efficacy of preoperative anti-tuberculosis therapy in older patients with spinal tuberculosis DOI Creative Commons
Shanshan Li,

Runrui Wu,

Mengru Feng

et al.

Frontiers in Cellular and Infection Microbiology, Journal Year: 2024, Volume and Issue: 14

Published: March 20, 2024

Spinal tuberculosis is a common extrapulmonary type that often secondary to pulmonary or systemic infections. Mycobacterium infection leads the balance of immune control and bacterial persistence. In this study, 64 patients were enrolled clinicopathological immunological characteristics different age groups analyzed. Anatomically, spinal in each group mostly occurred thoracic lumbar vertebrae. Imaging before preoperative anti-tuberculosis therapy showed proportion abscesses older was significantly lower than younger middle-aged groups. However, pathological examination surgical specimens higher other groups, there no difference granulomatous inflammation, caseous necrosis, inflammatory acute exudation, granulation tissue formation, fibrous hyperplasia. B cell number compared group, while T cells, CD4 + CD8 macrophages, lymphocytes, plasma NK cells did not differ. Meaningfully, we found IL-10 high expression TGF-β1 positive group. TNF-α, CD66b, IFN-γ, IL-6 expressions among three conclusion, are some differences imaging, pathological, features The may weaken their immunity treatment effectiveness.

Language: Английский

Citations

2
Ablation of Death-Associated Protein Kinase 1 Changes the Transcriptomic Profile and Alters Neural-Related Pathways in the Brain DOI Open Access
Ruomeng Li,

Shuai Zhi,

Guihua Lan

et al.

International Journal of Molecular Sciences, Journal Year: 2023, Volume and Issue: 24(7), P. 6542 - 6542

Published: March 31, 2023

Death-associated protein kinase 1 (DAPK1), a Ca2+/calmodulin-dependent serine/threonine kinase, mediates various neuronal functions, including cell death. Abnormal upregulation of DAPK1 is observed in human patients with neurological diseases, such as Alzheimer's disease (AD) and epilepsy. Ablation expression suppression activity attenuates neuropathology behavior impairments. However, whether regulates gene the brain, its profile implicated disorders, remains elusive. To reveal function pathogenic role diseases differential transcriptional profiling was performed brains knockout (DAPK1-KO) mice compared those wild-type (WT) by RNA sequencing. We showed significantly altered genes cerebral cortex, hippocampus, brain stem, cerebellum both male female DAPK1-KO to WT mice, respectively. The are multiple neural-related pathways, including: AD, Parkinson's (PD), Huntington's (HD), neurodegeneration, glutamatergic synapse, GABAergic synapse pathways. Moreover, our findings imply that potassium voltage-gated channel subfamily A member (Kcna1) may be involved modulation Our study provides insight into pathological regulatory networks new therapeutic strategies for treatment diseases.

Language: Английский

Citations

6
Analysis of the value of potential biomarker S100‐A8 protein in the diagnosis and pathogenesis of spinal tuberculosis DOI Creative Commons
Zhibo Ren,

Jinke Ji,

Caili Lou

et al.

JOR Spine, Journal Year: 2024, Volume and Issue: 7(2)

Published: April 10, 2024

Abstract Objectives The objective of this study is to evaluate the value S100‐A8 protein as a diagnostic marker for spinal tuberculosis and explore its role in potential pathogenesis (STB). Methods peripheral blood 100 patients admitted General Hospital Ningxia Medical University from September 2018 June 2021 were collected observation group, 30 healthy medical examiners control group. Three samples group three selected proteomics detection screening differential proteins. Kyoto Encyclopedia Genes (KEGG) was used enrich analyze related signaling pathways confirm target protein. serum expression levels proteins determined compared between two groups using enzyme‐linked immunosorbent assay (ELISA). Statistical methods STB. A macrophage model Mycobacterium infection constructed small interfering RNA investigate molecular mechanism Results has diagnosing (AUC = 0.931, p < 0.001), level (59.04 ± 19.37 ng/mL) significantly higher than that (43.16 10.07 ( 0.05). showed significant positive correlation with both CRP ESR values 0.01). Its AUCs combined bacteriological detection, T‐SPOT results, imaging, antacid staining pathological results 0.705 0.05), 0.754 0.01), 0.716 0.656 0.681 respectively. Lack leads decrease TLR4 IL‐17A infected macrophages. Conclusion differentially expressed individuals may be novel candidate biomarker diagnosis tuberculosis. feedback loop on S100‐A8‐TLR4‐IL‐17A axis play an important inflammatory

Language: Английский

Citations

1
CIDER: Counterfactual-Invariant Diffusion-based GNN Explainer for Causal Subgraph Inference DOI Creative Commons

Q. Zhang,

Chengshang Lyu, Lingxi Chen

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 9, 2024

Abstract Inferring causal links or subgraphs corresponding to a specific phenotype label based solely on measured data is an important yet challenging task, which also different from inferring nodes. While Graph Neural Network (GNN) Explainers have shown potential in subgraph identification, existing methods with GNN often offer associative rather than insights. This lack of transparency and explainability hinders our understanding their results underlying mechanisms. To address this issue, we propose novel method link/subgraph inference, called CIDER: Counterfactual-Invariant Diffusion-based ExplaineR, by implementing both counterfactual diffusion implementations. In other words, it model-agnostic task-agnostic framework for generating explanations counterfactual-invariant process, provides not only due implementation but reliable the process. Specifically, CIDER first formulated as inference task that generatively two distributions one another spurious subgraph. Then, enhance reliability, further model Thus, using distribution, can explicitly quantify contribution each phenotype/label manner, representing subgraph’s strength. From causality perspective, interventional method, traditional association studies observational approaches, reduce effects unobserved confounders. We evaluate synthetic real-world datasets, all demonstrate superiority over state-of-the-art methods.

Language: Английский

Citations

0