Peritoneal dialysis effluent biomarkers from a multi-omics and artificial intelligence perspective: advances and challenges

ABSTRACT

Long-term peritoneal dialysis (PD) treatment can lead to the destruction of peritoneal structure and function, which can lead to PD failure or even a poor prognosis. However, validated early biomarkers for patients undergoing PD are lacking. PD effluent (PDE) is rich in various biological components, such as nucleic acids, proteins, and metabolites, and is now an important source of noninvasive biomarkers for the dynamic monitoring of disease progression. In recent studies, a variety of histological techniques have provided unprecedented depth and breadth to PD biomarker research, and are becoming key tools in the early diagnosis, prognosis, and therapeutic monitoring of PD patients. Correspondingly, artificial intelligence (AI) approaches, which can flexibly handle data and excel at mining nonlinear and high-dimensional relationships in multimodal data, have moved from theory to practice. AI-based multi-omics analysis has not only greatly improved the understanding of the pathophysiological mechanisms of PD-associated fibrosis (PF) but has also contributed to the development of new biomarkers and novel targets. This review provides a comprehensive summary of recent advances in the development of PDE biomarkers using AI-based multi-omics approaches. We highlight the application of AI-based multi-omics techniques for early diagnosis, evaluation of peritoneal injury, assessment of peritoneal function, and prediction of prognosis. Finally, we discuss the challenges and limitations of PDE biomarkers from the perspectives of multi-omics and AI. In conclusion, AI-based multi-omics analysis holds great promise for the development of PDE biomarkers, which are expected to significantly improve the prognosis of PD patients and ultimately facilitate precision medicine.

INTRODUCTION

Globally, chronic kidney disease (CKD) affects >850 million people [1], with a median prevalence of 9.5% [2]. Approximately 5.3 million of these people have end-stage renal disease (ESRD) [3]. Peritoneal dialysis (PD) is one of the main renal replacement therapies used by patients with ESRD, accounting for 9% of all renal replacement therapies and 11% of all dialysis [4, 5]. Compared with hemodialysis, PD has several advantages, including home care, cost-effectiveness, less stress on the heart, and better preservation of residual renal function [4, 6]. However, patients receiving PD treatment often have to discontinue treatment prematurely because of complicating factors such as PD-associated fibrosis (PF), loss of peritoneal function or infectious peritonitis. Therefore, early monitoring of peritoneal function and complications of PD is particularly important.

Against the backdrop of an increasing demand for noninvasive monitoring, PD effluent (PDE) has garnered increasing attention because of its distinctive biological properties. As the material is directly available after solute exchange with capillaries during PD [7], PDE is rich in bioactive components. Therefore, PDE has become an important source of noninvasive biomarkers for the dynamic monitoring of disease progression in PD patients.

Here, we provide a systematic review of the research progress concerning PDE biomarkers that have been developed using multi-omics analysis and artificial intelligence (AI). On the basis of the achievements of research into these PDE biomarkers, it is expected that the precise prediction of complications, personalized treatment, and assessment of prognosis will be possible in the future, ultimately improving the quality of life and prognosis of patients with ESRD.

MECHANISM AND CURRENT CLINICAL APPLICATIONS OF PDE AS A BIOMARKER SOURCE

PDE is rich in a variety of components. These include a wide range of small solutes, macromolecules, detached cells, various substances secreted by peritoneal cells, and metabolic small molecules [8]. Dynamic changes in these components can accurately reflect the structural and functional state of the peritoneum, providing important biomarkers for clinical diagnosis and treatment [9].

Peritonitis is a serious complication in PD patients, and recurrent infections can lead to sclerosis of the peritoneum and eventual loss of ultrafiltration [10]. Currently, inflammatory factors in PDE, such as interleukin 6 (IL-6), carbohydrate antigen (CA125), and matrix metalloproteinase-8 (MMP-8), have been widely studied as early diagnostic markers of peritonitis in patients on PD [10–15]. In addition, studies have reported that calprotectin, glycoprotein 96 (GP96), lipopolysaccharide, cyclophilin A, neutrophil gelatinase-associated lipocalin (NGAL), and osteoprotegerin may also serve as biomarkers for the early diagnosis of peritonitis [16–24]. However, because PD patients often have multiple underlying diseases, many inflammatory markers are elevated in both systemic and local inflammation, making it difficult to distinguish the effects of local peritoneal inflammation from those of systemic inflammation.

PF is a complex pathological process. Currently, methods for diagnosing peritoneal fibrosis are shifting from invasive testing to noninvasive PDE marker evaluation, and common PDE biomarkers of PF include CA125, IL-1β, IL-18, decoy receptor 2 (DcR2), plasminogen activator inhibitor (PAI-1), MMP2, cc-chemokine ligand 18 (CCL18), HIF1A, and VEGF [25–35]. Although previous studies have identified several indicators of peritoneal function and structural changes in PDE, the lack of uniform laboratory standards for these indicators limits their use in clinical practice.

PDE plays an important role in the assessment of peritoneal transport function [36]. The traditional peritoneal equilibration test (PET) suffers from a lag time and a lack of dynamic monitoring capability [37]. Researchers have shown that aquaporin-1 (AQP1) and matrix metalloproteinase 2 (MMP2) have the potential to predict ultrafiltration function in patients with PD [29, 38–40]. Moreover, β-trace protein can be used to estimate the residual renal function of PD patients [41, 42]. Peritoneal transport is a dynamically changing process, and most current biomarker-based assessment methods detect it at a single point in time.

In recent years, the high-throughput omic technologies have developed rapidly in the field of PD. Next, we review the research progress in exploring PDE biomarkers on the basis of omic technologies (Table 1 and Fig. 1).

Table 1:

Overview of PDE biomarkers based on open omics research.

Targeting strategies	Targeting protein/drugs/compounds	Sample type	Research subjects	Mechanism/intervention effects	References
Transcriptomics	CD24	PDE	57 patients with PD	CD24 may serve as a potential biomarker to monitor peritoneal injury	[43]
	T-helper 2 cell	PDE	90 patients with PD	The mechanisms of adaptive immunity that promote the T-helper 2 cell response may trigger the processes that promote fibrosis	[109]
Single-cell transcriptomics	CCL5	PDE	6 patients with long-term PD and 6 patients with short-term PD	The inhibition of CCL5 may reduce immune cell infiltration and slow the fibrotic process	[46]
	Macrophages, T cells	PDE	6 patients with PD	Immune cells and PMC are associated with PF	[47]
	AQP1, MAP1LC3B	PDE	16 patients with PD	AQP1 can serve as an early diagnostic indicator for ultrafiltration failure and that MAP1LC3B may be potential therapeutic targets	[48]
	MiR-129–5p	PDE	16 patients with PD	miR-129–5p is significantly downregulated in PF	[51]
	MiR-21	PDE	230 patients with PD	miR-21 promotes fibrosis by inhibiting programmed cell death 4	[93]
	MiR-223, miR-31	PDE	107 patients with PD	Elevated levels of peritoneal miR-223 and reduced levels of miR-31 are useful predictors of bacterial infection	[53]
	miR-204–5p	Macrophage exosomes	Primary rat peritoneal macrophages	Exosomal miR-204–5p could be an effective target for treating PF	[56]
	Exosomal miRNA	PDE exosomes	3 patients with ultrafiltration failure and 3 control patients	MiR-125a-5p, miR-1273c-pc-3, miR-1277–5p, miR-132–3p, miR-296–3p and miR-708–5p are associated with PF	[55]
proteomics	Transferrin, catecholamines	PDE	9 patients with PD	Transferrin and catecholamine was could serve as a potent biomarker for the development of infectious peritonitis	[61]
	CCL11, CCL13, CCL19, CCL4, IL18	PDE	6 patients with PD	The chemokines and FLT3LG and IL18 could be used as early warning indicators.	[60]
	Intelectin-1, dermatopontin, gelsolin and retinol binding protein-4	PDE	10 patients with EPS and 18 patients in the control group	Intellectin-1 and dermatopontin are specifically elevated, providing a molecular basis for the early diagnosis of EPS	[58]
	EVs	PDE	14 patients with PD and mouse model	The percentage of ILK positive EVs in PDE correlated with peritoneal dysfunction and the degree of peritoneal damage	[62]
	EVs	PDE	8 patients with PD	EVs can serve as a source of biomarkers for PD patients	[65]
	AQP1	PDE exosomes	30 patients with PD	The presence of AQP1-containing exosomes can serve as an indicator of dialysis efficiency	[29]
	Glycoprotein 96	PDE exosomes	60 patients with PD	GP96 can serve as a potential biomarker for assessing the status of peritoneal inflammation and PSTR	[17]
	EVs	PDE exosomes	11 patients with PD	PDE-EV proteins cans serve as feasible biomarkers of PM alteration in PD patients	[37]
	protein tyrosine phosphatase 4A1 (PTP4A1)	PDE exosomes	12 patients with PD	PTP4A1 was the biomarker correlated with both the duration of PD treatment and the decline in peritoneal function	[67]
	Myeloperoxidase (MPO) and myeloperoxidase (HP)	PDE	15 patients with PDAP and 15 PD controls	MPO and HP may be potential diagnostic and therapeutic targets for PDAP	[110]
	Proteins	PDE	19 patients with APD	Proteomics methods can identify biomarkers	[111]
metabonomics	Alanine, creatinine, glucose, lactate	PDE	125 patients with PD	These metabolites can be used to distinguish between high and low PET types	[69]
	Citrulline and choline	PDE	19 patients with PD	Citrulline and choline can serve as new biomarkers for peritoneal function	[70]
	AlaGln	PDE	20 patients with PD	AlaGln supplementation indicate antioxidant effects	[71]
	kynurenine	PDE	8 patients with PD	kynurenine are associated with inflammation	[72]
microbiomics	Staphylococcus aureus and Klebsiella pneumoniae	anal swab	25 patients with PDAP and 45 PD controls	Staphylococcus aureus and Klebsiella pneumoniae are the most common pathogens causing PDAP	[82]
	Bacterial DNA fragments	PDE	143 patients with PDAP	The level of bacterial DNA fragments can predict recurrent peritonitis	[78]
	TMAO	PDE	144 patients with PD	TMAO have been found to significantly increase the risk of peritonitis.	[79]
	Gut flora	feces	30 patients with PD	There is an interaction between malnutrition and the gut flora in PD patients	[81]
	Lactobacillus and Prevotellaceae	feces	28 patients with PD and 29 patients with ESRD	Gut microbiome evaluation could aid early cognitive impairment diagnosis in patients undergoing PD	[23]
	Lactobacillus casei Zhang (LCZ)	PDE and feces	mouse model	LCZ has been shown to prevent PF	[80]
Other new technologies	Thrombospondin-1 (TSP1), collagen-13 (COL13), VEGFA	PDE	16 patients with PD	TSP1, COL13, and VEGFA are associated with peritoneal hypertranslocation and can serve as markers of decreased ultrafiltration function	[84]
	glycosylation	PDE	94 patients with PD	Altered glycosylation profiles in PDE associated with complications of peritonitis, inflammation and fibrosis	[86]

Figure 1:

PD Effluent Biomarkers from a Multiomic and AI. Abbreviations: IL-6, interleukin 6; MMP-8,matrix metalloproteinase 8; CA125, carbohydrate antigen 125; NGAL, neutrophil gelatinase-associated lipocalin; GP96, glycoprotein 96; LPS, lipopolysaccharide; CypA, cyclophillin A; OPG, osteoprotegerin; HDAC6, histone deacetylase 6; sICAM-1, soluble intercellular adhesion molecule-1; IGF2BP3, insulin-like growth factor 2 mRNA binding protein 3; eDcR2, effluent decoy receptor 2; miR-21, microRNA-21; MMP-2, matrix metalloprotein 2;AQP1, aquaporin-1;TGF-β, transforming growth factor beta; MMP9, matrix metalloprotein 9;CCL, CC-chemokine ligand; BTP, beta-trace protein; PAI-1, plasminogen activator inhibitor 1; MMP-7, matrix metalloprotein 7; eFDPs, effluent fibrin degradation products; NLRP3, NOD-like receptor protein 3; CTGF, connective tissue growth factor; Mφ, macrophages; DC, dendritic cells; AlaGln, alanyl-glutamine; IL 18,interleukin 18; TMAO, trimethylamine-N-oxide.

PERITONEAL FLUID BIOMARKERS BASED ON OPENOMICS STUDIES

Transcriptome

PDE is rich in cellular debris, extracellular vesicles, and free nucleic acids, which provide important information about gene expression. Transcriptomics can reflect the gene regulatory mechanisms involved in PD-related diseases; proteomics reflects protein expression and interactions and can provide guidance for diagnosis and treatment; and metabolomics can reflect changes in disease-related metabolic pathways. These multi-omics approaches provide a strong scientific basis for the accurate diagnosis and personalized treatment of PD patients.

Transcriptome sequencing

In recent years, transcriptome sequencing technology has enabled progress in studies related to the assessment of PD-related injury, PDAP, and the prognosis of patients undergoing PD [43–45]. Research has revealed that CD24 may serve as a potential biomarker to monitor peritoneal injury [43]; however, the findings have not been validated in clinical samples, and its translational value thus remains unconfirmed. In addition, some researchers have reported that an elevated ratio of CD14+ macrophages to CD1c+ dendritic cells may serve as a biomarker for predicting recurrent peritonitis and early catheter failure [44], and this phenomenon may be closely related to the core role of macrophages in immune responses. In another study, an analysis of integrated data from 12 publicly available transcriptomic databases revealed that the ACKR2–CCL2 axis was identified as a potentially pivotal actor in peritoneal functional deterioration [45]. Although transcriptomics has initially identified the above potential biomarkers in PD, it still faces many challenges overall, including the limited clinical sample size, lack of longitudinal follow-up data, and insufficient reproducibility and translatability of research results. Future multicenter prospective studies with large sample sizes are needed to strengthen the connection between basic research and clinical application to promote the clinical translation of transcriptomic findings.

Single-cell transcriptomics

Researchers have made some progress in the field of PF using advanced single-cell transcriptomic technology, which has greatly promoted the in-depth development of PDE biomarker research. Studies have demonstrated that the inhibition of CCL5 may reduce immune cell infiltration and slow the fibrotic process [46]. In addition, the development of a single-cell transcriptome atlas of the peritoneal microenvironment in long-term PD patients has further clarified the synergistic profibrotic mechanisms of macrophage, T-cell, and PMC subpopulations [47].

In addition, a study revealed that the downregulation of AQP1 can serve as an early diagnostic indicator for ultrafiltration failure and that autophagy-related epithelial-like mesothelial cell genes (e.g. MAP1LC3B) may be potential therapeutic targets [48]. Future efforts should validate the predictive efficacy of CCL5, AQP1, and others through longitudinal multicenter cohort studies. Integrating spatial multi-omics with targeted intervention research will advance the clinical translation of single-cell findings toward delaying fibrosis progression and reversing ultrafiltration failure, ultimately improving long-term outcomes for PD patients.

Noncoding RNA (miRNA, lncRNA, and circRNA) sequencing

MicroRNAs (miRNAs) are small, noncoding RNA molecules that are 19–25 nucleotides in length. Targeting mRNA at the posttranscriptional level through multiple signaling pathways has demonstrated core regulatory value in tumors, cardiovascular diseases, and fibrotic conditions [49–52]. In the field of PD, researchers have discovered a series of miRNA biomarkers in PDE through miRNA sequencing that have significant clinical application value for evaluating peritonitis and PF [50, 51, 53]. Research has revealed that elevated levels of miR-223 and reduced levels of miR-31 in the peritoneum serve as effective predictors of bacterial infection [53]. In the progression of PF, the expression of miR-129–5p is markedly downregulated, and its overexpression antagonizes the transforming growth factor beta1 (TGF-β1)-induced epithelial–mesenchymal transition (EMT) [51]. However, existing research remains largely confined to the discovery phase and single-center, small sample, cross-sectional expression profiling studies. The absence of standardized technical protocols and unclear reproducibility has hindered clinical translation and the advancement of multicenter validation.

In recent years, compared with free-floating miRNAs, miRNAs carried within extracellular vesicles have been demonstrated to exhibit superior stability owing to their encapsulation within phospholipid bilayer membranes. Imbalances in miRNA expression within extracellular vesicles are closely associated with tissue dysfunction and a variety of diseases [54]. In the field of PD, exosomal miRNA sequencing technology plays a significant role in evaluating peritoneal ultrafiltration function and PF [55–57]. A small sample study of patients experiencing ultrafiltration failure revealed multiple differentially expressed miRNAs via PDE exosome analysis. These may represent potential biomarkers for ultrafiltration failure; however, the study’s small sample size limits the statistical power of the differential miRNA analysis [55]. Furthermore, research has identified macrophage-derived extracellular vesicle miR-204–5p as an effective component for targeting PF [56], providing theoretical support for EV-miRNA therapeutic targets. Therefore, the importance of noncoding RNA sequencing in PD-related diseases should not be overlooked.

Proteomics

Detecting key protein levels in PDE has emerged as a reliable strategy for identifying potential biomarkers within PD outflow, with proteomic technology providing robust support for this approach [58–61]. Researchers discovered that the presence of transferrin in PDE could serve as a potent biomarker for the development of infectious peritonitis [61]. Furthermore, research has identified IL-6, IL-18, and FLT3LG as potential early biomarkers for predicting inflammatory status in PD patients [60]. Proteomic testing of PDE in patients with encapsulating peritoneal sclerosis (EPS) suggests that the levels of proteins such as intellectin-1 and dermatopontin are specifically elevated, providing a molecular basis for the early diagnosis of EPS [58]. While these proteins may serve as biomarkers for early diagnosis and intervention, further validation is needed. However, PD proteomics still faces limitations such as high sample heterogeneity, lack of standardized protocols, and limited cohort sizes. Future efforts should advance PDE proteomics from discovery to clinical application through multi-omics integration, AI-driven combination modeling of biomarkers, and longitudinal dynamic monitoring.

Recent research highlights that extracellular vesicles (EVs) in PDE carry substances such as proteins, DNA, RNA, and metabolites [62, 63], and can serve as biomarkers for assessing complication risks in PD patients [64, 65]. Moreover, proteomic analysis of PDE-EVs has demonstrated that PDE-EVs can reveal dynamic changes in signals of deteriorating peritoneal function earlier than conventional PET imaging [37]. In 2017, researchers reported the presence, isolation, and characterization of EVs in PDE from PD patients for the first time using proteomics [66]. Through proteomic analyses, numerous researchers have subsequently demonstrated that EVs are closely associated with PF, peritonitis, and impaired peritoneal transport function [17, 29, 37, 65, 67]. For example, the presence of AQP1-containing exosomes in PDE can serve as an indicator of dialysis efficiency [29]. Research has shown that GP96 in PDE-EVs can serve as a potential biomarker for assessing the status of peritoneal inflammation and the peritoneal solute transport rate in patients with PD [17]. Other studies have shown that protein tyrosine phosphatase 4A1 is closely associated with both the duration of PD treatment and the decline in peritoneal function [67]. Despite significant advances in extracellular vesicle proteomics, current research faces several challenges, including inconsistent standards for PDE-EV separation, limited sample sizes, high EV heterogeneity, dynamic variations in EV constituents, and a lack of large-scale clinical validation. Achieving the clinical translation of PDE-EVs necessitates establishing unified technical protocols.

Metabolomics

PDE is enriched with metabolites, including amino acids, sugars, lipids, and their derivatives. On the basis of metabolomic technology, biomarkers in PDE can be used to predict the occurrence of peritoneal dysfunction, the early prognosis, and the risk of PD-related complications [68].

In recent years, progress has been made in the application of metabolomic technology to PD, enabling the assessment of peritoneal transport functions, inflammation, and oxidative stress [69–72]. PDE metabolome analysis revealed marked alterations in the abundance of small-molecule metabolites during treatment, suggesting their potential influence in PD patients [71]. Recent research has revealed that metabolites such as alanine, creatinine, glucose, and lactic acid, in addition to small-molecule metabolites such as citrulline and choline, can be used to distinguish between high- and low-transport types [69, 70]. In addition, metabolomic changes associated with AlaGln supplementation indicate antioxidant effects, providing a basis for the correlation between metabolomics and personalized PD prescriptions [71]. Nontargeted metabolomic analyses have revealed that the levels of kynurenine and its metabolites in PD patients are associated with inflammation, oxidative stress and coronary artery disease [72], providing new biomarkers for the early diagnosis and treatment of PD-related complications. Therefore, metabolomics is important for assessing peritoneal function and personalized management.

However, metabolomics is limited by poor metabolite stability, low reproducibility, and the absence of absolute concentration measurements, making it difficult to establish clinical thresholds. Furthermore, biomarkers identified through metabolomics similarly lack multicenter, large-scale validation. Future efforts to advance metabolomics from discovery to clinical application may include the following: establishing standard operating procedures covering the entire workflow from sample collection to data submission; conducting prospective, multicenter cohort validation studies; elucidating metabolite functions through multi-omics integration; and developing low-cost, high-throughput, absolute quantitative targeted detection methods.

Microbiomics

The role of microbiomics in PF has rarely been discussed. However, it has recently been proposed that the gut flora may be involved in this process through various pathways, including the regulation of metabolites, activation of the inflammatory response and immune modulation [73]. Alterations in the gut flora associated with CKD lead to intestinal barrier dysfunction, allowing microorganisms, their debris, and toxins to enter the peritoneal cavity. This triggers intra-abdominal and systemic, chronic, low-grade inflammation, which promotes PF [74–76]. Therefore, gut microbiomics may offer a fresh approach for assessing complications and the prognosis in patients undergoing PD [77].

Gut microbes play regulatory roles at multiple levels in PD-related complications, affecting peritonitis, PF, nutrition, and cognitive function [23, 73, 78–82]. Microbiome sequencing has revealed that the level of bacterial DNA fragments and trimethylamine-N-oxide (TMAO) in PDE can predict recurrent peritonitis [78]. Moreover, TMAO are associated with accelerated progression of renal fibrosis [79, 83].

Furthermore, microbiome analysis is being increasingly used to study the composition and impact of gut microorganisms in fecal samples from PD patients. Microbiome sequencing has revealed interactions between malnutrition and the gut flora in PD patients, providing a theoretical basis for probiotic and prebiotic interventions [81]. Gut flora dysbiosis has also been linked to cognitive impairment [23]. Lactobacillus casei Zhang has been shown to prevent PF by remodeling the intestinal flora, activating PPAR-γ, and suppressing NF-κB-mediated inflammation [80]. These findings suggest new approaches for the early diagnosis of and probiotic interventions for PD-related complications.

However, PD microbiome research still faces multiple challenges. On the one hand, the widespread use of antibiotics may significantly disrupt microbial community structures, affecting the accuracy of the results. On the other hand, controlling contamination during sample collection and processing is difficult, particularly in PDE samples. In summary, although interactions along the gut–peritoneal axis may represent a key mechanism underlying PD treatment failure, the current research into microbiome-based biomarkers among PD patients remains extremely limited.

Other recent technologies

The emergence of new technologies, such as novel epigenomic, whole-genomic, mass spectrometry glycomic, and single-extracellular vesicle assays, in recent years is expected to provide more comprehensive technical support for the discovery of biomarkers for diseases such as PD-related inflammation and filtration dysfunction. Recently, a genome-wide microarray analysis revealed that proteins such as TSP1, COL13, and VEGFA can serve as markers of decreased ultrafiltration function [84]. Case reports on the use of whole-genome sequencing for diagnosing pathogens in PDAP have been published in clinical practice [85]. Investigators have used a glycomic approach involving mass spectrometry to analyze clinical samples from patients with PD [86]. Researchers have now applied single-cell extracellular vesicle analysis techniques to study biomarkers in oncology, cardiovascular disease, and neuroscience [87–89]. Given that PDE is rich in EVs, this technique could facilitate the further development of exosomal biomarkers in PDE. These latest technologies in the field of PD research are continuously being developed, providing new directions for the in-depth identification of key mechanisms and biomarkers of PD-related complications.

STRATEGIES AND INNOVATIVE APPROACHES TO INTEGRATE MULTI-OMICS TECHNOLOGIES

Studies of PDE biomarkers driven by the integration of multi-omics technologies

While single-omic technologies typically provide information on only one aspect of a biological system, integrating multi-omics technologies enables us to simultaneously gain a deeper understanding of the complex relationships between biomarkers and diseases at multiple levels [90]. Therefore, the application of multi-omics techniques in PD research can help overcome the limitations of single biomarkers by complementing them with those of different histological techniques [91] (Table 2).

Table 2:

Overview of PDE biomarkers based on the integrated research of multi-omic technologies.

Targeting strategies	Targeting protein/drugs/compounds	Sample type	Research subjects	Mechanism/intervention effects	Multi-omics	References
multi-omics	microRNA	PDE	12 patients with PD and mouse model	MicroRNAs in the peritoneum are potential therapeutic targets for PF	Single-cell transcriptome and metabolome	[92]
	transforming growth factor-β1 (TGF-β1)	PDE	55 patients with PD and mouse model	Transforming growth factor beta 1 (TGF-β1) promotes the development of PF	Proteomics and single-cell RNA sequencing	[62]
	αB-crystallin	PDE	11 patients with PD	αB-crystallin plays a role in PD-associated angiogenesis and fibrosis	Transcriptomes and Proteomics	[94]
	Gut microbial metabolites	——	——	The role of key endogenous metabolites originating from the gut microbiome in PD patients	Microbiomics and metabolomics	[95]
	Gut microorganisms	Serum and feces	72 patients with PD , 13 patients with ESRD and 13 health volunteers	Specific changes in the gut microecology and metabolic system in patients undergoing PD	Microbiomics and metabolomics	[96]

The integration of multi-omics technologies in biomarker research is highly important, and some recent advances have been made in the field of PD, playing a vital role in the diagnosis and treatment of PF [62, 92–94]. Recently, studies integrating single-cell transcriptomic and metabolomic data have revealed the pathological mechanisms of the MMT and PF driven by high glycolysis in mesothelial cells, suggesting that metabolic enzymes and associated microRNAs may serve as potential intervention targets [92]. Through an integrated analysis of miRNAomic and transcriptomic data, a study revealed that microRNA-21 (miR-21) promotes fibrosis by suppressing genes such as programmed cell death 4 (PDCD4), further confirming its significance as a biomarker [94]. However, these findings require external validation to substantiate its role. Furthermore, by using single-cell RNA sequencing and exosomal proteomic techniques, researchers have revealed that blocking EV secretion or knocking down integrin ligase kinase (ILK) expression significantly inhibits fibroblast activation and PF progression [62]. Through integrated proteomic and transcriptomic analyses, several studies have also revealed the central role of αB-crystallin in promoting PD-associated angiogenesis and fibrosis [94]. Furthermore, studies have investigated the role of key endogenous metabolites originating from the gut microbiome in PD patients using combined microbiome and metabolome analyses [95, 96].

However, research into PD biomarkers driven by multi-omics technology remains in its infancy. Most studies to date constitute either foundational research or single-center clinical investigations with small sample sizes, with shortcomings such as low reproducibility, high risk of bias, and limited external validity. Moreover, omics research invariably carries inherent methodological challenges, including batch effects, variations in sample processing, and biases in reference databases and annotations. In conclusion, these strategies lay the foundation for a comprehensive understanding of disease pathogenesis, as well as the development of new biomarkers and therapies for clinical use. Future prospective large-scale clinical cohort studies are needed to validate the efficacy of these strategies.

AI-driven moderation network building among multi-omics data

AI technologies, such as machine learning and deep learning, are gradually being integrated into various aspects of PD research (Table 3). Many researchers are currently using AI algorithms to process and integrate rapidly increasing amounts of multi-omics data, analyze the underlying molecular networks involved in disease progression, and improve the accuracy of disease detection [97–99].

Table 3:

Overview of PDE biomarkers based on the integrated research of AI and multi-omic technologies.

Targeting strategies	Targeting protein/drugs/compounds	Research subjects	AI algorithms and validation	Mechanism/intervention effects	Multi-omics	References
AI + multi-omics	TSP1, collagen-13 , VEGFA	16 patients with PD 921 patients with PD	Endurance prediction: Extra Trees Regressor Cause of PD end: Linear Discriminant Analysis (LDA) Validation: Independent external validation set (n = 32)	Omics technologies alongside AI algorithms can be used to develop biomarkers.	Whole-Genome RNA Microarray Analysis and AI	[84] [104]
	microRNA	142 patients with PD.	Multiple logistic regression, AdaBoost, Decision tree, Gradient tree boosting, random forest Validation: Internal validation set (n = 15)	Combining transcriptomics with AI could help with the early diagnosis of EPS.	Transcriptomes and AI	[105]
	gut microor- ganisms	101 patients with PD	Cox proportional hazards model Validation: A time-dependent ROC curves	A decrease in the diversity of PDE flora is has been found to be independently associated with PD technique failure.	Microbiomics and AI	[106]
	SOCS1	86 patients with PD	XGBoost, random forest, and support vector machine (SVM) Validation: Clinical samples (n = 86) and animal models (n = 24)	SOCS1 is a potential marker of peritoneal fibrosis.	Transcriptomes and AI	[107]
	——	——	——	Data-driven strategies have potential to improve the outcomes of diabetes management for PD patients.	Multi-omics and AI	[108]

The integration of AI and multi-omics is actively advancing the development of precision medicine in PD. In the field of PD, recent studies have employed omic technologies combined with AI algorithms to develop predictive models for PD failure [100], hospitalization duration, and the risk of cardiovascular mortality and all-cause mortality in PD patients [101–103]. For instance, the MAUXI model employs a combination of Extra Trees and linear discriminant analysis (LDA) to predict PD technology survival and failure based on molecular characteristics associated with MMT in PDE. Its validation encompasses both internal validation and external validation using an independent dataset from another hospital, thereby enhancing the robustness of the model [84, 104]. In addition, researchers have shown that combining transcriptomics with AI could aid in the early diagnosis of EPS and provide new directions for its treatment [105]. This study combined ratios of five miRNAs with clinical characteristics, employing five algorithms including AdaBoost and random forests to construct predictive models demonstrating favorable diagnostic performance. However, the study lacks detailed elaboration on parameter settings for each algorithm, feature selection mechanisms, and model interpretability. Furthermore, despite employing 5-fold cross-validation, the model was validated only within a single center without external validation in multicenter or independent cohorts, raising questions about external validity. Future research should include interpretability analysis and validation of the robustness and clinical applicability of the model in larger, multicenter cohorts. Concurrently, practical challenges for clinical deployment—including standardization of testing, data heterogeneity, long-term stability, and ethical considerations—must be addressed.

In the field of microbiology combined with AI, reduced bacterial diversity in PD fluid has been demonstrated to be independently associated with PD technology failure [106]. This study constructed a predictive score based on a Cox proportional hazards regression model, integrating three indicators: age, history of diabetes, and gut microbial diversity [106]. However, the method of model construction in this study did not employ modern AI or machine learning algorithms and instead relied on traditional statistical modeling. While the Cox model offers the advantage of high interpretability, it is less capable than machine learning approaches of handling higher-dimensional data and identifying novel biomarkers. Recent research has identified potential biomarkers for PF, including suppressor of cytokine signaling 1 (SOCS1), by integrating three machine learning algorithms—XGBoost, random forest, and support vector machine (SVM)—based on GEO transcriptomic data [107]. Future work requires larger-scale, multicenter prospective studies to facilitate the transition of such AI models into clinical practice.

Similarly, personalized nutritional management of diabetes in PD patients has been achieved through the use of multi-omics and AI [108]. AI can also assist clinicians by automating data interpretation, improving treatment planning and enhancing patient education. Personalized, data-driven strategies have great potential to improve the comprehensive management and prognostic outcomes of PD patients [108]. However, its clinical implementation still requires overcoming multiple hurdles, including large-scale validation, equipment consumables, and cost considerations. In summary, although AI-driven multi-omics models demonstrate broad application prospects in the field of PD, their clinical rollout urgently requires large-scale, multicenter, prospective validation. Furthermore, addressing algorithm transparency, model stability, and feasibility assessments for implementation is essential for advancing clinical translation.

DISCUSSION

Research into PDE biomarkers is moving toward intelligent strategies, systematization, and greater precision. In the future, AI models will efficiently screen and validate PDE markers, thereby improving the efficiency of discovery. By combining PDE biomarkers with clinical indicator prediction models, we can construct systems for disease prediction and treatment assessment. Combining multi-omics methods will identify more accurate PDE biomarkers.

However, this pathway still faces multiple bottlenecks. Regarding omics data, high multi-omics heterogeneity, inconsistent experimental protocols and quality control standards, and difficulties in data integration pose significant challenges. Furthermore, the lack of consensus on PD fluid collection, storage, and processing protocols exacerbates batch effects. At the AI level, batch effects and small sample sizes increase the risk of overfitting in AI models, reducing their generalizability and limiting omics research. Batch effects may render data from different experimental groups incomparable, thereby impacting biomarker discovery and validation. Small sample sizes may lead to insufficient statistical power, making it difficult to detect genuine biological differences, which directly contributes to the risk of overfitting. Regarding clinical translation, existing research predominantly remains confined to laboratory-based fundamental studies and single-center clinical trials with small sample sizes. The absence of large-scale, multicenter, prospective cohort studies for external validation constrains the clinical applicability and dissemination value of these models. Moreover, AI applications face ethical controversies, privacy concerns, and risks of data breaches. Robust data protection measures and strict adherence to regulatory standards are essential for mitigating risks and safeguarding patient data security. Concurrently, challenges persist, including inconsistent data quality, model bias, insufficient explainability, high maintenance costs, and gaps in adaptive regulation.

Future efforts must focus on establishing standardized technical frameworks, strengthening interdisciplinary collaboration, and creating data-sharing mechanisms to overcome current technical limitations and advance the clinical translation of PDE biomarker research.

SUMMARY AND OUTLOOK

This review summarizes the latest findings on PDE biomarkers based on transcriptomic, proteomic, metabolomic, and other multi-omics techniques combined with AI technology to integrate multidimensional data to provide promising noninvasive biomarkers and predictive models for PD in terms of complications and prognosis. Recent research has transcended the limitations of single-molecule detection, progressing toward multi-omics integration and AI-based predictive models. Through the continuous optimization of technical approaches, these biomarkers will significantly improve the prognosis of PD patients. Future work should focus on developing and validating AI models with enhanced clinical interpretability and generalizability, advancing through phased implementation. Integrating multi-omics data into dynamic predictive models will improve biomarker temporal sensitivity and personalized accuracy. Large-scale, multicenter prospective cohort validation will ensure robustness. Interpretable AI techniques can be combined to elucidate the biological significance of key feature genes, thereby enhancing clinical credibility. Additionally, the integration of AI models with electronic health record systems can be promoted to achieve seamless translation from laboratory to bedside applications. Ultimately, the integration of multi-omics technologies with AI will further increase both the efficiency of biomarker discovery and the clinical value of these biomarkers, providing patients with more precise diagnostic and therapeutic solutions.

FUNDING

This work was supported by the National Natural Science Foundation of China (82200838), Chongqing Science Foundation Project (CSTB2023NSCQ-MSX0584), Chongqing Technology Innovation Project (2023QNXM007) and National Clinical Research Funding (2023-NHLHCRF-YS-01, 2022-173ZD-112).

DATA AVAILABILITY STATEMENT

No new data were generated or analyzed in support of this research.

CONFLICT OF INTEREST STATEMENT

None declared.