The study, published in Frontiers in Bioinformatics by Yumeng Liu and Gang Luo of the Department of Pediatrics at Liaoning Maternal and Child Health Hospital, took aim at a long-standing gap in sepsis medicine. Childhood sepsis cannot be regarded as a mere miniature version of adult sepsis. Children undergo continuous development of their immune systems, and the inflammatory response, alterations in organ function, and clinical diagnostic criteria following infection all exhibit age-related characteristics. Consequently, candidate markers identified in adult cohorts cannot be directly applied to the diagnostic context observed in children. By screening and validating molecular markers specifically within child cohorts, the researchers sought diagnostic insights more aligned with the clinical and biological characteristics of pediatric sepsis.

The scientific premise of the work rests on oxidative stress, a state that arises when reactive oxygen species, or ROS, are produced in amounts greater than the body can neutralize with antioxidants. This imbalance drives mitochondrial dysfunction, which has been identified as a central pathophysiological mechanism underlying organ failure in sepsis, both in experimental models and in clinical patients. In a landmark study of 28 severe sepsis patients during the first 24 hours of admission, Brealey and colleagues observed that mitochondrial dysfunction, along with antioxidant depletion and impaired energy metabolism, were key drivers of multiple organ failure. Recent clinical studies in pediatric populations have demonstrated that affected children exhibit significant alterations in oxidative stress markers and antioxidant capacity, indicating that disruption of the oxidative-antioxidant balance may play a critical role in the pathophysiology of pediatric sepsis. Excessive ROS generation also triggers the MAPK and PI3K/Akt signaling pathways, elevates NF-κB levels, and consequently induces the expression of pro-inflammatory genes, and elevated NF-κB in sepsis patients serves as a stronger predictor of disease severity than conventional inflammatory markers.

To hunt down the molecular fingerprints of this oxidative storm, the researchers mined five open-access transcriptomic datasets from the Gene Expression Omnibus database, all comprising peripheral whole-blood samples from children with sepsis and healthy controls, all generated on the same GPL570 microarray platform to ensure comparability. Two datasets, GSE26440 and GSE26378, were merged to form the training cohort, while three independent datasets, GSE13904, GSE9692, and GSE8121, were combined into a validation cohort. The team corrected batch effects using the empirical Bayes ComBat method, verified the integration with principal component analysis, and then applied differential expression analysis with the limma package, using thresholds of |log2 fold change| greater than 0.585 and an adjusted p-value below 0.05. This rigorous pipeline yielded 396 significant differentially expressed genes, including 290 upregulated and 106 downregulated genes, and after intersecting these with a curated list of 817 oxidative stress-associated genes from prior literature, 36 differentially expressed oxidative stress-related genes emerged as the study’s raw material.

Functional enrichment analysis revealed what these 36 genes were doing. Gene Ontology analysis highlighted their roles in regulating inflammatory responses and reactive oxygen species metabolism, while Kyoto Encyclopedia of Genes and Genomes pathway analysis identified nine pathways implicated in sepsis progression, including neutrophil extracellular trap formation, the Toll-like receptor and NOD-like receptor signaling pathways that orchestrate innate immunity, and the regulation of inflammatory mediators affecting TRP channels. Neutrophil extracellular traps, or NETs, the DNA-based web-like structures that neutrophils deploy to capture and kill pathogens, were flagged as particularly critical in sepsis pathogenesis. NET formation relies on ROS generated by NADPH oxidase, and in sepsis an excessive immune response drives uncontrolled NET production that aggravates lung injury. Gene set enrichment analysis further showed that the oxidative stress phenotypic score was significantly higher in the sepsis cohort than in controls, with upregulation of ROS metabolic processes and sepsis-associated pathways in the high-score group.

The methodological heart of the study was a two-stage computational gauntlet. First, weighted gene co-expression network analysis, or WGCNA, was applied to the top 15 percent most variable genes in the training cohort. At a soft-thresholding power of beta equal to 8, the network closely approximated a scale-free topology, and the dynamic tree-cut algorithm resolved 14 distinct co-expression modules. One module, dubbed turquoise, showed a striking positive correlation with sepsis, with a Spearman coefficient of 0.65 and a p-value of 4.57 times 10 to the power of minus 29, marking its member genes as collectively pivotal in disease progression. Second, six machine learning algorithms, Bayesian analysis, bagged trees, LASSO regression, random forest, learning vector quantization, and Boruta, were each set loose on the oxidative stress-related differentially expressed genes. Only genes identified by all six methods, a threshold of six out of six, were retained as robust candidates. This consensus rule ensured that the screening results were not dependent on any single algorithm, and it ultimately distilled the field down to seven candidate diagnostic genes: DDAH2, GADD45A, HP, MAPK14, PGD, RETN, and S100A9.

From these seven candidates, the researchers constructed a protein-protein interaction network using the STRING database and excluded DDAH2 for low connectivity, leaving six hub genes: GADD45A, HP, MAPK14, PGD, RETN, and S100A9. Each of these genes encodes a protein with a plausible mechanistic link to sepsis biology. MAPK14, better known as p38 alpha, is a central kinase in inflammatory signaling cascades. S100A9 is an alarmin released abundantly by activated neutrophils and monocytes. HP is haptoglobin, an acute-phase protein that binds free hemoglobin. PGD encodes an enzyme in the pentose phosphate pathway central to cellular antioxidant defense. GADD45A participates in DNA repair and stress responses, and RETN, or resistin, is an inflammatory adipokine. Expression levels of all six hub genes were significantly elevated in septic children compared with controls, and receiver operating characteristic analysis revealed area under the curve, or AUC, values of 0.951, 0.982, 0.964, 0.944, 0.962, and 0.982 in the training data, and between 0.908 and 0.977 in the independent validation cohort, a level of diagnostic performance that few single biomarkers in clinical use today can match.

The team did not stop at individual genes. Seven machine learning classifiers, including random forest, support vector machine, LogitBoost, k-nearest neighbors, Naive Bayes, AdaBoost, and CancerClass, were trained on the pooled training cohort with five-fold cross-validation repeated ten times, and then evaluated on the untouched validation cohort. The k-nearest neighbors model, which classifies samples by their similarity to neighboring expression profiles, achieved the highest AUC in validation and was selected as the final predictive model. A nomogram built from the six hub genes pushed diagnostic accuracy even further, reaching an AUC of 0.994. The researchers also performed immune cell deconvolution using CIBERSORT, estimating the relative proportions of 22 immune cell types in peripheral blood. The analysis revealed marked alterations in the sepsis cohort: M0 macrophages, M2 macrophages, monocytes, activated mast cells, and neutrophils were significantly elevated, while follicular helper T cells, CD8 T cells, CD4 memory T cells, resting dendritic cells, and activated NK cells were markedly reduced. Most hub genes were positively correlated with neutrophils and negatively linked with resting dendritic cells, suggesting that oxidative stress-related genes may modulate the abundance and activity of distinct immune cell subsets in the immune microenvironment.

Perhaps the most clinically tantalizing result came from the drug-discovery arm of the study. The researchers queried the Drug-Gene Interaction database and the Connectivity Map to identify small molecules predicted to downregulate the hub genes, then performed blind molecular docking using the CB-Dock2 platform. The MAPK14 protein demonstrated strong binding affinities for doxycycline and ralimetinib mesylate, with AutoDock Vina scores of minus 8.4 and minus 8.1 kilocalories per mole respectively, while the S100A9 protein exhibited significant binding affinity for tasquinimod at minus 7.0 kilocalories per mole. Doxycycline, an antibiotic already in widespread clinical use, engaged MAPK14 through multiple ionic bonds, hydrophobic forces, and hydrogen bonds, while ralimetinib mesylate, a p38 MAPK inhibitor, and tasquinimod, an experimental immunomodulator, engaged their targets through complementary interaction networks. These findings position MAPK14 and S100A9 as promising therapeutic targets and open up new treatment options that could be explored for repurposing in pediatric sepsis.

The authors are careful to frame their findings as a foundation rather than a finished clinical tool. The analysis is based entirely on publicly available retrospective datasets, and heterogeneity across cohorts in sample sources, platforms, and clinical annotations may still influence performance estimates. The immune cell deconvolution relied on a single algorithm and reflects peripheral blood composition rather than tissue-level infiltration, and the molecular docking results suggest potential binding but do not demonstrate therapeutic efficacy in patients. The ceRNA network of 102 nodes and 160 edges, which implicates 89 long non-coding RNAs and 10 microRNAs in the post-transcriptional regulation of GADD45A, MAPK14, and PGD, remains exploratory pending experimental validation. Still, the study’s significance is hard to overstate. Compared with conventional biomarkers such as procalcitonin, C-reactive protein, and lactate, this six-gene oxidative stress signature may more sensitively reflect early molecular alterations associated with immune dysregulation, providing complementary diagnostic information precisely when conventional indicators are inconclusive. If prospective, multicenter validation confirms these results, the era of catching pediatric sepsis in its first critical hours, before organ failure takes hold, may move one decisive step closer to reality.

Subject of Research: Oxidative stress-related biomarkers and machine learning diagnostic models for pediatric sepsis

Article Title: Identification and validation of biomarkers related to oxidative stress in pediatric sepsis utilizing WGCNA and multiple machine learning algorithms

Article References: Liu, Y., & Luo, G. (2026). Identification and validation of biomarkers related to oxidative stress in pediatric sepsis utilizing WGCNA and multiple machine learning algorithms. Frontiers in Bioinformatics, 6, Article 1745381. https://doi.org/10.3389/fbinf.2026.1745381

Image Credits: AI Generated

DOI: 10.3389/fbinf.2026.1745381

Keywords: pediatric sepsis, oxidative stress, biomarkers, machine learning, WGCNA, GADD45A, MAPK14, S100A9, molecular docking, immune cell deconvolution, diagnostic model, reactive oxygen species