Biomarker Discovery Research Articles (Page 1)

Chronic kidney disease (CKD) represents a major and expanding global health challenge, with prevalence rising due to aging populations, diabetes, hypertension, and environmental factors. Conventional risk assessment tools such as the CKD Epidemiology Collaboration equation and the Kidney Failure Risk Equation are limited in precision, generalizability, and their ability to identify rapid progressors in early stages. This review examines the transformative role of artificial intelligence (AI), encompassing machine learning, deep learning, natural language processing, and multimodal data integration, in improving CKD detection, progression prediction, and personalized management. Drawing on recent evidence, we highlight AI's capacity to process high‑dimensional data from electronic health records, imaging, omics, and wearable devices, achieving area under the curve values of 0.85-0.96 for predicting outcomes, such as end‑stage kidney disease and therapeutic response. Key applications include early CKD screening using gradient boosting and long short‑term memory networks, biomarker discovery through multi‑omics fusion, and precision phenotyping to guide targeted interventions such as sodium-glucose cotransporter‑2 inhibitor therapy and optimized dialysis initiation. Persistent challenges algorithmic bias, data privacy, interpretability, and regulatory compliance-necessitate strategies such as federated learning, explainable AI, and ethically guided, equitable implementation.

The Proteograph Product Suite, a multiplexed nanoparticle (NP) protein corona-based workflow, substantially improves the depth of detection of proteins by mass spectrometry (MS) by compressing the dynamic range of protein abundances. Here, we evaluate its quantitative performance and suitability for large-scale studies. Using multispecies spike-in experiments, we assessed fold change accuracy, linearity, precision, and the lower limit of quantification (LLOQ) across multiple MS platforms. Combined with the Orbitrap Astral MS, the Proteograph XT assay enabled identification of more than 7,000 plasma proteins. In mixed-species dilution experiments, fold change accuracy was preserved, with Proteograph quantifying 3.5 times more proteins than the Neat plasma workflow at the same fold change error threshold. Similar accuracy was observed with the Orbitrap Exploris 480 MS, and we also demonstrate that different proteome backgrounds do not impact the accuracy. Data produced with NPs from the four distinct NP batches (each supporting >100,000 assays) showed only a 4% increase in protein intensity CV across batches. Together, these results demonstrate that the Proteograph Product Suite provides depth as well as quantitative accuracy and precision to power new biomarker discovery and biological understanding in population-scale plasma proteomics cohorts.

Mycotoxin exposure contributes to adverse human health outcomes, however, data regarding validated human biomarkers of exposure are lacking. This study presents an integrated framework for the biomarker discovery and toxicokinetic characterization of mycotoxin in humans. The aim of the study is to identify new biomarkers, determine their toxicokinetic (TK) properties, and build an integrated data analysis workflow using machine learning (ML), whilst focusing on non- and minimally-invasive sampling strategies. Following sample collection and chemical analysis, obtained datasets are used for the computation of ML models. Probability-based techniques are employed to calculate specific boundaries in the multidimensional space and, in parallel, ML classification methodologies are evaluated to scrutinize controls from intervened volunteers. Furthermore, multivariate regression models are computed to study the correlation of potential biomarkers with mycotoxin dosages. Once biomarkers have been identified, data are fit using Bayesian methods to a population-TK model to estimate key parameters related to absorption, distribution, metabolism, and excretion. This standardized framework allows the scientific community to identify and validate new mycotoxin biomarkers and related ADME-properties in both a precise and accurate manner. Although we developed the proposed trial for various different mycotoxins, due to ethical considerations, focus was set towards IARC group III-classified mycotoxins.

Liquid biopsy noninvasively characterizes diseases by analyzing biomarker proteins in biofluids, which provide valuable insights into physiological and pathological processes. In this study, we conducted an analysis of the differences between atypical endometrial hyperplasia or endometrial cancer (AH/EC) patients after fertility-sparing treatment with different RNA-Seq-based endometrial receptivity test (rsERT) results ("receptive" versus "pre-receptive"), to investigate the proteomic connections among tissue, serum, and urine samples. Samples of endometrial tissue, serum, and urine from 40 rsERT "pre-receptive" and 10 rsERT "receptive" patients were analyzed for proteomic profiling. We integrated differentially expressed proteins from three sample types to investigate endometrial receptor (ER)-related molecular changes. Our findings indicated that both serum and urine proteomes can serve as indicators of functional changes in endometrial tissue. In serum, proteins associated with cholesterol metabolism, immune response, and coagulation exhibited a differential expression. In urine, proteins related to immune function and metabolic processes demonstrated varying levels of expression. The differentially expressed proteins in both serum and urine were associated with the immune response and metabolism. In conclusion, biofluids serve as a reflection of functional changes in endometrial tissue, which will facilitate a deeper understanding of endometrial receptivity and the discovery of potential clinical biomarkers.

Mass spectrometry imaging has emerged as a pivotal tool in spatial metabolomics, yet its reliance on the imzML format poses critical challenges in data storage, transmission, and computational efficiency. While imzML ensures cross-platform compatibility, its lower compressed binary architecture results in large file sizes and high parsing overhead, hindering cloud-based analysis and real-time visualization. This study introduces an enhanced Aird compression format optimized for spatial metabolomics through two innovations: (1) a dynamic combinatorial compression algorithm for integer-based encoding of m/z and intensity data; (2) a coordinate-separation storage strategy for rapid spatial indexing. Experimental validation on 47 public data sets demonstrated significant performance gains. Compared to imzML, Aird achieved a 70% reduction in storage footprint (mean compression ratio: 30.89%) while maintaining near-lossless data precision (F1-score = 99.75% at 0.1 ppm m/z tolerance). For high-precision-controlled data sets, Aird accelerated loading speeds by 13-fold in MZmine. The Aird format overcomes crucial bottlenecks in spatial metabolomics by harmonizing storage efficiency, computational speed, and analytical precision, reducing I/O latency for large cohorts. By achieving near-native feature detection accuracy, Aird establishes a robust infrastructure for translational applications, including disease biomarker discovery and pharmacokinetic imaging.

Honey bees (Apis mellifera) are vital pollinators essential for maintaining ecosystem stability and global food production, but they face escalating threats from pathogens, agrochemicals, and climate change. Although proteomics has advanced our understanding of bee physiology, single-omics approaches are insufficient to capture the complexity of colony health. This review highlights the rise of integrative multiomics frameworks─combining proteomics, metabolomics, and lipidomics─with artificial intelligence (AI)-based strategies to decode molecular resilience in bees. We summarize recent advances in omics technologies, including spatial and single-cell platforms, mass spectrometry innovations, and customized computational pipelines. Furthermore, we highlight how AI-enhanced multiomics integration facilitates biomarker discovery, elucidates regulatory networks, especially in nonmodel organisms like honey bees. Emerging computational methods such as deep learning, graph neural networks, and multilayer network models offer predictive, scalable, and interpretable insights. Despite challenges like limited sample input and cross-omics heterogeneity, the convergence of omics and machine learning represents a transformative paradigm for decoding complex biological systems. These integrative approaches offer not only a deeper molecular understanding of bee biology but also generalizable frameworks for systems biology in other ecologically relevant species.

XGBoost, a gradient boosting algorithm, is widely recognized for its efficiency and robustness in multiclass classification tasks. Metabolomics serves as a powerful tool for biomarker discovery; however, metabolic biomarkers associated with the progression from chronic hepatitis B (CHB) to liver cirrhosis (LC) to hepatocellular carcinoma (HCC), as well as those related to treatment effects in HCC (HCCAT), remain unclear. In this study, an XGBoost-based machine learning approach combined with mass spectrometry was used to analyze the metabolic profiles of 30 healthy controls (HC), 29 CHB patients, 30 LC patients, 30 HCC patients, and 30 HCCAT patients. Biomarker screening was conducted through three comparative analyses: (1) HC, CHB, LC, HCC, and HCCAT; (2) HC, CHB, LC, and HCC; and (3) HC, HCC, and HCCAT. A total of 17 metabolic biomarkers were identified, among which nine had not been previously associated with HBV-related liver diseases. Notably, a potential biomarker panel composed of eicosenoic acid, dihydromorphine, cysteine, acetic acid, sitosterol, and hypoxanthine showed promise for disease prognosis and therapeutic evaluation. These findings highlight the great potential of integrating metabolomics with machine learning to identify novel metabolic biomarkers related to HBV-associated liver disease progression and treatment response.

Heart failure represents the terminal stage of diverse cardiovascular disorders and continues to impose a heavy global burden despite advances in therapy. Beyond classical neurohormonal and hemodynamic pathways, recent studies underscore the central role of non-coding RNAs in orchestrating epigenetic remodeling that drives hypertrophy, fibrosis, and cardiomyopathy. In this review, we synthesize evidence into an integrative “ncRNAs–epigenetics–cardiomyopathy” working model that connects upstream non-coding RNAs regulation with chromatin dynamics and downstream pathological remodeling. We integrate evidence showing how microRNAs, long non-coding RNAs, and circular RNAs reshape transcriptional networks through interactions with DNA methylation, histone, and RNA modifications. Differential non-coding RNAs signatures across heart failure phenotypes, comorbidities, and complications are further highlighted, underscoring their potential utility in patient stratification and biomarker discovery. We also evaluate therapeutic frontiers that extend beyond single-target interventions toward multi-layered approaches, including antisense oligonucleotides, CRISPR/dCas9-mediated epigenome editing, and exosome- or nanoparticle-based delivery systems. Although translational barriers remain considerable, especially in terms of specificity, safety, and clinical validation, these strategies illustrate the potential of targeting the ncRNAs–epigenetic axis to advance precision medicine in heart failure.

Letter to editor on "deep learning facilitated discovery of prognosis biomarkers and their ligands to improve liver cancer treatment".

Background: While chronic pathological substrates of sudden cardiac death (SCD) are well-known, the molecular triggers that predispose individuals to fatal arrhythmias that day remain undefined, posing a barrier to near-term prevention. We recently reported the first transcriptome study of human myocardium sampled at autopsy-confirmed SCD to demonstrate upregulation of active fibrosis and selected channel dysregulation as a vulnerable substrate for lethal arrhythmias via candidate gene approach. Here, the study of comprehensive proteomes in autopsy-confirmed SCD samples enables unbiased discovery of novel pro-arrhythmic pathways and biomarkers. Hypothesis: The paired myocardium and circulating blood of autopsy-defined SCDs exhibit distinct proteomes compared to non-cardiac sudden deaths and trauma deaths that reflect their vulnerable myocardial state and its associated circulating biomarkers in the hours to days before SCD. Aims: Define the proteome of coupled myocardial tissue, neat serum, and enriched serum of autopsy-confirmed sudden deaths. Methods: We conducted unbiased, quantitative proteomics via LC-MS on paired LV myocardium and serum samples from 13 arrhythmic SCDs (5 acute MI, 3 CAD, 3 DCM, 2 LVH) vs. 7 non-arrhythmic sudden deaths (occult overdose, myocardial rupture, and traumas) to define the arrhythmic substrate. Results: Comprehensive proteomics resolved 5,979 proteins from myocardial tissue and 6,013 proteins from serum, representing a 3-fold increase in proteome depth vs. previous reports. Among these ~6,000 proteins, 2,088 were upregulated and 662 downregulated in Arrhythmic vs. Non-arrhythmic deaths. Arrhythmic myocardium exhibited 3-fold higher fibrosis-associated proteins (POSTN, LUM, FMOD, COL1A1) and 2-fold lower contraction-related proteins (TNNC1, TNNI3, MYLK3, MYL4). Our novel enrichment method resolved 3,645 proteins otherwise undetectable in the neat fraction (POSTN and COL1A1). Reactome analysis of overlapping differentially expressed proteins in myocardium and enriched serum revealed coordinated upregulation of extracellular matrix remodeling, metabolic dysfunction, and immune response. Conclusion: The proteome of myocardium and serum at SCD reveals systemic and cardiac-specific molecular changes which together may exacerbate existing fibrosis and contribute to electrical instability and acutely vulnerable arrhythmogenic substrate. These findings provide candidate biomarkers for near-term risk assessment of fatal arrhythmias.

Background: Right ventricular (RV) function is the principal determinant of exercise capacity and survival in patients with pulmonary arterial hypertension (PAH). Despite its clinical significance, the molecular mechanisms driving maladaptive RV remodeling and fibrosis remain incompletely understood, limiting therapeutic options and biomarker discovery. Emerging evidence highlights the regulatory roles of tRNA-derived small RNAs (tsRNAs) in cardiovascular disease, yet their involvement in PAH-associated RV fibrosis has not been elucidated. Methods: We performed high-throughput sequencing to profile tsRNA expression in plasma from PAH patients and RV tissues from monocrotaline (MCT) rats and established two rat models of PAH (MCT and SU5416 combined with hypoxia). Functional effects of tRF-1:32-Glu-TTC-2-M2, a significantly downregulated tsRNA, were assessed by AAV9-mediated overexpression in vivo and mimic transfection in primary cardiac fibroblasts in vitro. Right heart function was evaluated by echocardiography and cardiac magnetic resonance imaging. Protein interactions were investigated by chromatin isolation by RNA purification coupled with mass spectrometry, RNA pull-down, and RNA immunoprecipitation assays. Rescue experiments validated downstream targeted protein. Results: tRF-1:32-Glu-TTC-2-M2 was markedly reduced in the RV but not lung tissues of PAH patients and rat models, inversely correlating with RV fibrosis and dysfunction. Overexpression of tRF-1:32-Glu-TTC-2-M2 significantly ameliorated RV fibrosis, improved RV function, and enhanced survival independently of pulmonary vascular remodeling. Mechanistically, tRF-1:32-Glu-TTC-2-M2 directly interacted with prolyl 4-hydroxylase subunit alpha-2 (P4HA2), inhibiting its protein expression and enzymatic activity, resulting in decreased collagen synthesis and cardiac fibroblast activation. Restoration of P4HA2 partially reversed these antifibrotic effects. Furthermore, plasma levels of tRF-1:32-Glu-TTC-2-M2 correlated with clinical indices of RV function, highlighting its potential as a diagnostic and prognostic biomarker. Conclusions: Our findings reveal a novel tsRNA-mediated regulatory axis in PAH-associated RV fibrosis, where tRF-1:32-Glu-TTC-2-M2 attenuates maladaptive remodeling through inhibition of P4HA2. This study identifies tRF-1:32-Glu-TTC-2-M2 as both a promising therapeutic target and biomarker for PAH, paving the way for RV-directed precision medicine strategies.

Untargeted metabolomics has emerged as a transformative approach in sports nutrition research, offering an unbiased means to characterize the complex biochemical responses to exercise, training, and dietary interventions. Unlike targeted assays restricted to predefined metabolites, untargeted strategies capture broad metabolic perturbations across lipid, carbohydrate, amino acid, and nucleotide pathways, enabling the discovery of novel biomarkers and unanticipated physiological mechanisms. This review critically evaluates the design and application of untargeted metabolomic pipelines in the context of exercise and nutrition science, from pre-analytical sample handling and analytical platforms such as NMR, LC-MS, and GC-MS, to data processing using tools like XCMS, MZmine, and MS-DIAL, and subsequent statistical and bioinformatic interpretation. Key applications include delineating acute metabolic shocks induced by endurance exercise, identifying athlete-specific metabolic phenotypes shaped by chronic training, and assessing the impact of nutritional interventions such as fruit intake, amino acid supplementation, or polyphenol-rich foods on exercise recovery and oxidative stress. The integration of metabolomics with other omics, particularly microbiome metagenomics and lipidomics, highlights the potential for systems-level insights into host-microbe-diet interactions. Nonetheless, significant challenges remain, including the reproducibility of findings, difficulties in metabolite identification, and the translational gap between large datasets and actionable nutritional strategies. By synthesizing current strengths, limitations, and controversies, this review emphasizes that the future of sports metabolomics lies in methodological standardization, multi-omics integration, and validation of candidate biomarkers in independent cohorts. Collectively, these efforts position untargeted metabolomics as a cornerstone for advancing precision nutrition and personalized performance monitoring in athletes.

Protein biomarkers are routinely measured for disease diagnosis and prognosis in clinical laboratories. Since most assays focus on protein quantity, information about proteoforms is often not acquired. Proteoforms of a protein represent the complex integration of genetic polymorphism, alternative splicing of RNA transcripts, and post-translational modifications (PTMs) on the amino-acid backbone. A detailed analysis of the post-translationally modified proteoforms (PTMPs), which are influenced by pathophysiological conditions, may lead to more precise diagnosis and prognosis. This article first discusses the methodologies used to accurately detect and characterize PTMPs, i.e., immunoassays, electrophoresis, chromatography, and intact and proteolysis-aided mass spectrometry techniques. Then it reviews specific examples of PTMP biomarkers that have been successfully translated from biomarker discovery to clinical use. The examples include β2-transferrin for cerebrospinal fluid leak diagnosis, phosphorylated tau proteoforms for Alzheimer disease diagnosis, and fucosylated alpha-fetoprotein for hepatocellular carcinoma prognosis. In addition, the article provides prospective views of novel analytical technologies and promising new PTMP biomarkers entering clinical practice. In summary, PTMs are controlled by biochemical processes to modulate the functions of proteins by expanding their chemical diversity. PTM alterations in proteins can be indicators for pathophysiological conditions. Advances in analytical technologies are deepening our understanding of PTMPs and paving the way for their translation to clinical use. As research continues to discover the clinical meaning of PTMP biomarkers, they are poised to become valuable additions to the clinical testing menu for precision medicine.

This review summarizes recent advances in applying metabolomics to autoimmune hepatitis (AIH). AIH is a chronic liver disease characterized by immune-mediated hepatocellular injury, with complex pathogenesis involving genetic, immunological, and environmental factors. Metabolomics, a system-wide approach analyzing small molecule metabolites, offers potential in early diagnosis, prognosis, and therapeutic evaluation of AIH. Current studies identify alterations in amino acid, lipid, carbohydrate, and bile acid metabolism, as well as changes in the gut microbiome and specific metabolite markers that distinguish AIH from other liver diseases. Techniques such as liquid chromatography-mass spectrometry (LC–MS), and bioinformatics facilitate biomarker discovery and enhance understanding of disease mechanisms. Despite challenges such as standardization and data integration, metabolomics holds promise for developing personalized treatment strategies and advancing disease management. Future prospects include combining multi-omics approaches, large-scale cohort studies, and artificial intelligence (AI)-based data analysis to deepen insights into AIH pathology and improve clinical outcomes.

Amiodarone is one of the most effective and commonly prescribed class III antiarrhythmic agents. However, its pharmacokinetics may be impaired and toxicity may be aggravated by metabolic disorders. The review presents data showing that dyslipidemia, obesity, and type 2 diabetes mellitus significantly affect pharmacokinetics, pharmacodynamics and toxic profile of amiodarone. These conditions alter the drug’s binding to lipoproteins, the activity of cytochrome P450- dependent monooxygenases and tissue distribution, thereby increasing the risk of accumulation and toxic effects. Chronic heart failure further impairs drug metabolism and contributes to multisystem toxicity. The current clinical guidelines do not adequately address these critical aspects. Therefore, more rigorous monitoring of these patients and their plasma drug concentration (on individual basis) is suggested, along with a generally more cautious approach to amiodarone use. Future perspectives include prospective clinical trials on amiodarone, physiologically based pharmacokinetic modeling for personalized dosing, based on body mass index, lipid profile, and comorbidities, pharmacogenomic studies (cytochrome P450 polymorphisms), as well as biomarker discovery for drug toxicity prediction.

Introduction: Arterial calcification contributes significantly to cardiovascular morbidity, yet its metabolic mechanisms remain poorly understood. This study aimed to characterize the spatial metabolic landscape of vascular calcification using matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI) in both human and animal tissues. Methods: Fresh calcified coronary arteries from cadaver hearts (n=8) and femoral arteries from a porcine model of arterial calcification (n=10, including calcified and controls) were snap-frozen and underwent untargeted metabolomic profiling using a UV-laser MALDI source (Spectroglyph LLC) coupled with a Thermo Q Exactive HF-X Orbitrap MS. Adjacent tissue sections were stained with Von Kossa to identify calcified regions (VK+), which were annotated and co-registered with MALDI-MSI datasets to correlate spatial metabolic features with mineralization. Metabolite analysis (>1550 features) was performed using QuPath and SCiLS, with statistical comparison by PCA and Pearson correlation (p<0.001). Results: In human coronary arteries, PCA plot indicated distinct metabolic profile between VK- and VK+ samples (Fig.1A). Spatial metabolomics analysis revealed a distinct set of 79 metabolites adjacent to calcified regions, primarily glycerophospholipids (38), sphingolipids (6), amino acid derivatives (5), fatty acids (3), and nucleosides&nucleotides (2), with 49 significantly elevated and spatially localized to calcified regions. In the porcine model, difference between VK+ and VK- regions was found (Fig.1B) and a robust panel of 90 metabolites was significantly associated with calcification (|r|>0.5), the majority of which were amino acid derivatives (32), fatty acids (6), nucleosides and nucleosides (5), and glycerophospholipids (4), consistent with human coronary artery results. Notably, 22 metabolites, including adenosine monophosphate, pyruvate, and phosphatidylinositol (20:2/16:0), were associated with pathways such as glycolysis and purine metabolism, mirrored the metabolic alterations observed in human tissues, supporting the translational relevance of the model. Conclusion: This study demonstrates the power of MALDI-MSI for high-resolution spatial metabolomic profiling of arterial calcification. The identified metabolite signatures provide new insights into the pathobiology of arterial calcification and may inform future therapeutic strategies targeting vascular mineralization and early biomarker discovery.

Plasma proteomics technologies are advancing rapidly, offering new opportunities for biomarker discovery and precision medicine. Direct comparisons of available technologies are needed to understand how platform selection affects downstream findings. We compared the performance of a peptide fractionation-based mass spectrometry method (HiRIEF LC-MS/MS) and the Olink Explore 3072 proximity extension assays on 88 plasma samples, analyzing 1129 proteins with both methods. The platforms exhibited complementary proteome coverage, high precision, and concordance in estimating sex differences in protein levels. Quantitative agreement between platforms was moderate (median correlation 0.59, interquartile range 0.33-0.75), mainly influenced by technical factors. Finally, we present a publicly available tool for peptide-level analysis of platform agreement and demonstrate its utility in clarifying cross-platform discrepancies in protein and proteoform measurements. Our findings provide insights for platform selection and study design, and highlight the value of combining mass spectrometry and affinity-based approaches for more comprehensive and reliable plasma proteome profiling.

Recent advances in mass spectrometry-based proteomics have enabled increasingly precise characterization of protein modifications in clinical specimens. Among these, glycosylation is one of the most structurally complex and biologically informative post-translational modifications, reflecting cellular differentiation and disease states. Ohashi et al. (J. Biochem. 2024; 175: 561-572) performed a site-specific N-glycosylation analysis of LAMP1 in breast cancer tissue samples, demonstrating the feasibility of targeted glycoproteomics in patient-derived specimens and revealing tumor-associated glycoform heterogeneity. Their study exemplifies how focusing on a single glycoprotein target can provide detailed insight into disease-specific glycan remodeling within the tumor microenvironment. In this commentary, I discuss the significance of such targeted approaches in the broader context of clinical glycoproteomics and highlight their potential contribution to cancer biomarker discovery and precision medicine. Continued integration of glycoproteomic data with genomic and clinical information is expected to further advance our understanding of tumor biology and therapeutic response.

BackgroundKiwifruit is widely recognized for its nutritional value and health benefits, yet reliable and objective methods for assessing kiwifruit intake in populations remain limited.ObjectiveThis study aimed to identify urinary biomarkers of kiwifruit intake and develop an optimal biomarker panel for differentiating consumers within days.MethodsA randomized, controlled, crossover dietary intervention was conducted among 17 healthy volunteers. The intervention included four phases: run-in, single‐exposure, repeat‐exposure, and follow-up. Urine samples at multiple time-point and fruit samples were prepared and analyzed using untargeted metabolomics via dual-column ultra-high-performance liquid chromatography-mass spectrometry (UHPLC-MS). Candidate biomarkers were identified through a systematic statistical strategy on kinetic profiles within 24 h, and annotated for potential fruit-derived origin through spectral matching. Machine learning algorithms were employed to establish an optimal biomarker panel for assessing kiwifruit intake under habitual diet conditions.ResultsTwenty-three urinary metabolites showed significantly elevated kinetic profiles, among which 15 were matched to compounds detected in the original fruit or in vitro digestion samples. These metabolites mainly included polyphenol-related metabolites and plant-derived amino acid derivatives. The excretion of many metabolites turned to be delayed compared to those typically observed for other fruits. For example, 2-isopropylmalic acid usually peaked in urine or blood within 6 h of consuming other fruits, but in our study urinary level at 24 h was much higher compared to 6 h. Most of the selected candidates are not specific to kiwifruit based on existing literature, such as hippuric acid. In this regard, an XGBoost algorithm-based model using 7 metabolites achieved substantial discriminative performance (accuracy = 0.88) in predicting kiwifruit intake within two days.ConclusionsThis study identified potential biomarkers of kiwifruit and developed a prediction model that may differentiate consumers. Further validation is necessary to confirm the reliability and generalizability of our findings.Trial registrationChinese Clinical Trial Registry, ChiCTR2100048279. Registered on July 5, 2021.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12937-025-01230-5.

BackgroundBreast cancer (BC) continues to be a predominant cause of cancer-related deaths among women globally, highlighting the complexity of the disease, its propensity for metastasis, and resistance to conventional therapies. The discovery of clinical biomarkers and therapeutic targets is crucial for advancing BC treatment strategies.MethodsWe conducted a Transcriptome-Wide Association Study (TWAS) and Weighted Gene Co-expression Network Analysis (WGCNA) to pinpoint genes associated with BC, with GSTM4 emerging as a candidate tumor suppressor. To further elucidate GSTM4’s role, we analyzed its mRNA and protein expression using multiple databases. The prognostic significance of GSTM4 was evaluated using KM-Plotter tools, while its epigenetic profile was examined through GSCA. Protein–Protein Interaction (PPI) networks were constructed with GeneMANIA and STRING to explore GSTM4’s functional interactions. The ssGSEA algorithm and TIMER were employed to link GSTM4 with the immune context of BC. Single-cell RNA sequencing from GSE148673 was analyzed to investigate GSTM4’s influence on the tumor immune microenvironment. In vitro experiments were designed to assess GSTM4’s impact on BC cell behavior.ResultsGSTM4 expression is diminished in BC tissues and is positively linked to better patient outcomes. Epigenetic studies indicate that GSTM4 silencing may partly result from promoter deletion mutation. The PPI network and enrichment analysis are consistent with GSTM4’s multifaceted role in glutathione metabolism, detoxification, antioxidant defense, and oxidoreductase reactions. GSTM4 expression shows a negative correlation with immune checkpoint gene expression and is associated with an enhanced antitumor immune response in BC as well as increased sensitivity to anticancer drugs. Single-cell RNA sequencing analysis suggests a association of GSTM4 with the prevention of epithelial transformation, the mitigation of BC cell malignancy, and the promotion of tumor microenvironment interactions. In vitro studies confirm GSTM4’s inhibitory effects on BC cell proliferation, invasion, and stemness.ConclusionIntegrated analyses identify GSTM4 as a potential tumor suppressor and prognostic biomarker in BC, suggesting its promise for advancing therapies.Supplementary InformationThe online version contains supplementary material available at 10.1007/s12672-025-03850-z.