Machine learning in lupus nephritis: bridging prediction models and clinical decision-making towards personalized nephrology

Background: Lupus nephritis (LN) is one of the most severe manifestations of systemic lupus erythematosus (SLE), affecting up to 65% of patients and contributing significantly to morbidity and mortality. The heterogeneous clinical course of LN—characterized by alternating flares and remissions—stems from complex immunological, genetic, endocrine, and environmental factors. Current management strategies rely on immunosuppressants and corticosteroids, yet predicting disease progression, treatment response, and relapse risk remains challenging.

Objective: This review synthesizes current evidence on the use of machine learning (ML) models for predicting, diagnosing, and monitoring LN, emphasizing their translational potential to improve clinical decision-making and enable personalized nephrology.

Methods: A narrative synthesis was conducted of studies published between 2015 and April 2024, identified through PubMed using the terms (“lupus nephritis” OR “LN”) AND (“machine learning” OR “artificial intelligence” OR “deep learning”). Eligible studies included those applying ML models to LN for diagnosis, histological classification, flare prediction, treatment response, or prognosis.

Results: We identified diverse ML approaches—including logistic regression, decision trees, random forests, support vector machines, neural networks, gradient boosting, and clustering—applied to multimodal data sources (clinical, laboratory, imaging, histopathology, and omics). These models demonstrated high performance in tasks such as non-invasive histology classification (AUC up to 0.98), flare prediction, and individualized risk stratification. Integration with big data frameworks enhanced the identification of molecular drivers, improved prognostic accuracy, and facilitated remote patient monitoring. However, model development in LN remains limited by small datasets, lack of external validation, and heterogeneous outcome definitions.

Conclusion: ML models have the potential to transform LN management by enabling earlier flare detection, personalized treatment strategies, and non-invasive disease monitoring. To achieve clinical integration, future research must prioritize robust validation, interoperability with electronic health records, and transparent model interpretability. Bridging the gap between computational performance and real-world application could substantially improve outcomes and quality of life for LN patients.

1 Introduction

Lupus nephritis (LN) is one of the most severe manifestations of systemic lupus erythematosus (SLE), affecting up to 65% of patients during the disease (1, 2). Its clinical course is heterogeneous, characterized by alternating periods of exacerbation and remission, and influenced by a complex interplay of immunological, endocrine, genetic, and environmental factors (3–5). Renal involvement ranges from subclinical disease to end-stage renal disease (ESRD), in which a generalized pro-inflammatory state accelerates renal function decline and significantly worsens patient survival (6).

There is currently no definitive cure for SLE or LN. Since the 1950s, standard treatment has aimed to induce remission, suppress disease activity, reduce symptoms, preserve renal function, and maintain remission (7). Although therapeutic regimens have evolved over time (induction vs. maintenance strategies), they typically combine an immunosuppressant with an intermediate-acting glucocorticoid to prevent persistent inflammation, irreversible renal damage, and progression to ESRD (8).

Multiple factors influence LN progression, including dysregulation of autoantibody production, poor adherence to therapy, excessive sun exposure (9), and socioeconomic disadvantages (10). However, these variables alone have limited predictive value for anticipating disease flares or renal deterioration (5). In this regard, machine learning (ML) algorithms offer the ability to incorporate multiple clinical and biological variables simultaneously, detect hidden patterns, and generate predictive models with greater accuracy (2).

The application of ML to LN monitoring provides several potential benefits. These include timely interventions to prevent disease progression and complications (11–15), the development of personalized follow-up strategies based on patient-specific characteristics and trajectories (14–17), and the ability to identify high-risk patients who may require closer surveillance. Moreover, ML models can predict the likelihood of flares by analyzing historical and longitudinal data, enabling clinicians to implement preventive measures such as therapy adjustments or lifestyle modifications (6, 12).

Another major advantage of ML is its capacity to integrate diverse data sources—including clinical variables, imaging, genomics, and patient-reported outcomes—thus offering a more comprehensive view of disease dynamics (16, 17). In addition, ML-based monitoring systems allow for remote, real-time patient follow-up, improving convenience, facilitating early intervention, and reducing the burden on healthcare resources (18). Taken together, these features position ML as a promising non-invasive complement to renal biopsy, capable of supporting clinical decision-making with predictive models that encompass a wide range of patient factors (4).

Considering the above, the guiding research question of this review is: Can machine learning algorithms meaningfully improve the prediction and monitoring of lupus nephritis, thereby enhancing clinical decision-making and advancing personalized treatment?

2 Methodology

This review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines. Although the synthesis is presented in a narrative format, all stages of the review—search, selection, extraction, and synthesis—were conducted systematically to ensure transparency and reproducibility. A systematic narrative review was designed to identify, analyze, and synthesize studies applying machine learning (ML) techniques to lupus nephritis (LN). The review focused on how ML models have been used to improve diagnosis, prognosis, monitoring, and prediction of therapeutic response in patients with LN.

A comprehensive literature search was conducted in PubMed, Scopus, and Embase for publications between January 2015 and July 2025, combining controlled vocabulary and free-text terms such as “machine learning,” “artificial intelligence,” “deep learning,” and “lupus nephritis.” Boolean operators (AND, OR) were applied to optimize the search results. Additionally, reference lists of the included articles were manually screened to identify further studies not captured in the initial search.

Predefined inclusion and exclusion criteria were applied to maintain methodological rigor. Eligible studies included original peer-reviewed research articles, systematic reviews, or meta-analyses published in English that applied ML techniques to LN for diagnostic, prognostic, monitoring, or treatment-response purposes. Case reports, editorials, and conference abstracts without full text were excluded, as were studies that did not explicitly employ ML algorithms in a clinical or translational context.

Title and abstract screening were conducted, followed by full-text evaluation of potentially eligible studies. Data extraction was performed using a standardized template including information on study design, cohort characteristics, ML algorithm type, data modalities (clinical, imaging, histopathology, omics), and reported performance metrics (accuracy, sensitivity, specificity, AUC). All extracted data were independently cross-checked by the reviewers to ensure completeness and reliability.

The evidence synthesis was performed narratively and organized according to the main clinical outcomes: diagnostic and histological classification, risk stratification and prognosis, prediction of therapeutic response, and longitudinal monitoring. A critical comparative analysis was conducted across algorithms, data types, and validation strategies to highlight methodological advances, limitations, and emerging trends.

To enhance methodological consistency and traceability, this review integrated four LLM-based agents—Planner, Researcher, Analyzer, and Documenter— which were employed to support and structure the review process. These agents are components of an Intelligent Multi-Agent Assistant, developed within the framework of a doctoral research project by one of the investigators. The Planner Agent defined the workflow and research milestones; the Researcher Agent assisted in query generation and metadata extraction; the Analyzer Agent facilitated thematic clustering and the identification of trends across studies; and the Documenter Agent ensured coherence, version control, and proper formatting of the extracted information (19). All AI-assisted operations were manually verified by the authors to ensure alignment with PRISMA standards, data integrity, and the scientific objectives of the review.

3 Results of literature review

3.1 Diagnosis

The application of machine learning (ML) in the diagnosis of lupus nephritis (LN) has significantly advanced in recent years, enabling non-invasive classification, earlier detection, and more precise differentiation of renal involvement. ML algorithms have been trained using multimodal data—clinical, serological, histopathological, and imaging—to complement conventional biomarkers such as anti-dsDNA antibodies, complement (C3/C4) levels, and proteinuria.

Recent work by Wang et al. (1) introduced a clinically oriented ML pipeline using ensemble classifiers, including XGBoost and random forest models, to assist in the diagnosis of LN. The model integrated standard clinical parameters and achieved an average AUC > 0.95 for both ROC and PRC curves, outperforming conventional diagnostic markers. Similarly, Chen et al. (2) developed an ML-based flare prediction system for LN using dynamic clinical and serological variables, demonstrating high sensitivity and specificity in distinguishing active from quiescent disease.

Deep learning approaches have revolutionized histopathological assessment in LN. Zheng et al. (6) trained a convolutional neural network (CNN) for automated glomerular lesion recognition in digitized biopsy slides, obtaining accuracies exceeding 90% compared with pathologist-based scoring. Moreover, a deep learning model by Huang et al. (20) predicted renal flare in LN from longitudinal multivariable datasets, emphasizing the feasibility of continuous, image-integrated diagnostics that surpass static biopsy evaluations. These findings support the role of ML as a complementary diagnostic tool, particularly in reducing observer variability and enhancing reproducibility of histological grading.

Non-invasive imaging modalities have also benefited from ML integration. In a recent ultrasound-based study, Qin et al. (4) built a radiomics-driven ML model for evaluating LN activity, reporting an AUC = 0.95 in training and AUC = 0.77 in test cohorts. Such radiomics-enhanced approaches combine structural and textural ultrasound features with serological indicators to distinguish active renal inflammation from chronic damage. Additionally, biomarker-driven diagnostic frameworks, such as those proposed by Guo et al. (5), combine combinatorial biosensor signals and ML classifiers to provide point-of-care diagnostic support, expanding accessibility to precision diagnostics in LN.

3.2 Risk stratification and prognosis

Machine learning models have been increasingly applied to predict disease activity, renal flare, and progression in lupus nephritis (LN), integrating multidimensional data to enable individualized risk stratification. Huang et al. (20) developed a deep learning model based on multivariable time-series data from 1,694 patients with biopsy-proven LN. Using a long short-term memory (LSTM) network with an attention mechanism, the model incorporated 59 clinical, immunologic, and therapeutic features and achieved a C-index of 0.897 in the validation set. Temporal variation in feature importance highlighted serum albumin, complement C3, and urinary protein as key predictors of renal flare (20) (Table 1).

Table 1

Table 1. Performance metrics of machine learning models applied to prediction and prognosis in lupus nephritis (2016–2024).

Stojanowski et al. (11) employed an artificial neural network (ANN) with a multilayer perceptron architecture to predict complete renal remission in 58 patients with proliferative LN. The algorithm reached an accuracy of 91.7% and an area under the ROC curve (AUC) of 0.94, outperforming conventional regression-based prognostic models (11). The same study also reported that integrating routine laboratory and clinical variables improved early risk discrimination for patients unlikely to achieve remission.

Mou et al. (21) applied 12 machine-learning algorithms and non-negative matrix factorization (NMF) to transcriptomic immune-gene datasets to identify prognostic molecular signatures in LN. Their model generated robust predictive performance with external-validation AUCs exceeding 0.90 and identified hub immune-related genes strongly correlated with glomerular filtration rate, proteinuria, and serum creatinine (21).

Tang et al. (22) used multivariate linear regression combined with feature-selection procedures to estimate acute and chronic histopathologic indices from clinical variables in 202 patients with biopsy-confirmed LN. The models achieved R² = 0.77 for chronic-index prediction and Q² = 0.52 for acute-index prediction, providing early quantitative evidence that clinical and biochemical data can approximate histologic activity and chronicity (22).

3.3 Treatment response prediction

Machine learning (ML) approaches have been increasingly implemented to predict therapeutic response in lupus nephritis (LN), particularly in the context of immunosuppressive and biological therapies. Lee et al. (23) developed a hybrid predictive framework combining transcriptomic profiling with ML classifiers to estimate treatment response after the first renal flare. The model integrated hub-gene expression signatures and topological features of regulatory networks derived from public microarray datasets. Using a random forest-based pipeline, it achieved an accuracy of 0.91, precision of 0.89, and AUC of 0.94, identifying STAT1, IRF7, and IFI44L as dominant predictive genes associated with response to mycophenolate mofetil (23).

Wang et al. (24) performed a bioinformatic and ML-driven analysis to detect driver genes influencing treatment sensitivity in LN. The study used an integrated dataset of 316 patients from GEO repositories, combining LASSO regression, support-vector machine-recursive feature elimination (SVM-RFE), and random-forest modeling. Cross-validation demonstrated consistent accuracy (AUC = 0.93) in identifying candidate driver genes (CCL5, CXCL9, ISG15) linked to responsiveness to corticosteroid and cyclophosphamide therapy (24).

Chen et al. (2) implemented a supervised ML algorithm using dynamic clinical variables to stratify flare risk during maintenance therapy. Although primarily designed for prognostic monitoring, the model’s discriminative capacity (AUC = 0.89, sensitivity = 0.84, specificity = 0.86) provided indirect evidence of its ability to forecast response to treatment intensification (2).

An et al. (8) described a precision-medicine-oriented ML strategy for individualized therapy optimization in LN, integrating immunologic biomarkers, baseline histologic indices, and therapeutic regimens. Their gradient-boosting ensemble model yielded an AUC = 0.88 and predicted achievement of partial or complete remission at 12 months. Feature-importance ranking highlighted baseline proteinuria, anti-dsDNA titers, and complement C3 levels as the most informative predictors (8) (Table 2).

Table 2

Table 2. Summary of key studies applying machine learning in lupus nephritis (2016–2024).

3.4 Monitoring and big data approaches

Recent advances in data integration and computational modeling have significantly expanded the scope of lupus nephritis (LN) monitoring beyond traditional laboratory and histopathological assessments. The combination of machine learning (ML) with big data analytics has enabled the development of systems capable of detecting disease activity, predicting flare risk, and optimizing therapeutic response through multimodal datasets encompassing clinical, biochemical, imaging, and molecular information.

In the clinical setting, Tang et al. (25) developed a serum biomarker miniarray supported by ML algorithms to continuously track disease activity and flare risk in LN patients, demonstrating enhanced sensitivity compared with standard serological markers such as anti-dsDNA and complement levels. Similarly, Deng et al. (26) implemented a natural language processing (NLP) framework within electronic health records (EHRs) to identify LN phenotypes using structured and unstructured data, improving diagnostic precision and temporal disease tracking across healthcare systems. These approaches illustrate the transition toward data-driven monitoring strategies capable of detecting subtle patterns of renal inflammation before clinical manifestation.

The incorporation of deep learning (DL) into multimodal monitoring platforms has also shown promising results. Li et al. (27) proposed a DL-based system integrating retinal imaging with clinical variables to detect systemic lupus and its renal complications, highlighting the feasibility of remote and non-invasive screening methods that capture systemic and microvascular alterations. Likewise, Zhan et al. (17) emphasized that ML frameworks leveraging multi-omic and EHR data fusion are redefining patient stratification and longitudinal tracking in SLE, setting a methodological foundation for precision nephrology.

Population-based approaches have emerged to complement individualized monitoring. Izadi et al. (28) developed and validated a risk scoring system trained on large clinical cohorts to identify LN cases within general SLE populations, showing robust discriminatory power when applied across multiethnic datasets. Additionally, the use of high-throughput sequencing and omics-level data mining has provided a molecular basis for tracking disease heterogeneity. Studies integrating transcriptomic and epigenomic profiles through ML pipelines have elucidated signatures of renal injury progression and therapeutic response dynamics (29, 30).

Parallel efforts in nephrology have explored the convergence of big data infrastructure with ML-assisted analytics. Gomathi and Narayani (31) pioneered the integration of cloud-based big data pipelines for autoimmune disease prediction, outlining a scalable computational framework applicable to LN. More recently, Agrawal et al. (32) described the foundational architecture that supports distributed data storage and retrieval essential for ML-driven analysis of large nephrological datasets. Databases such as UK Biobank and the Lupus Foundation of America’s ALPHA project have enabled large-scale aggregation of clinical, imaging, and sociodemographic data, which has fueled algorithmic refinement and validation of predictive models in LN (33).

Collectively, these monitoring strategies rely on continuous data acquisition, integration of EHR-derived features, and multi-omic analytics, establishing the groundwork for proactive surveillance and personalized disease management within the emerging paradigm of data-intensive lupus nephritis care (Table 3).

Table 3

Table 3. Machine learning and deep learning applications for non-invasive diagnosis and monitoring of lupus nephritis (2016–2024).

4 Discussion

The reviewed studies exhibit notably high discriminative performance, with many models achieving area under the curve (AUC) values above 0.85 in internal validation cohorts. However, model bias, interpretability, and generalizability remain major challenges that condition clinical translation, as performance often declines when models are applied to external or multiethnic populations (34). For example, in a study of proliferative LN prediction, all models achieved AUCs exceeding 0.80, and a ridge regression variant attained AUC = 0.953 in the training cohort and maintained values above 0.80 in held-out testing sets, demonstrating strong classification capacity even across different algorithmic approaches (35). In diagnostic work, Wang et al. reported an AUC of 0.995 for their optimized XGBoost pipeline using selected features such as proteinuria, lupus anticoagulant, and RBP, outperforming traditional biomarkers like anti-dsDNA and complement levels (1). Such results underscore the strength of machine learning in integrating multidimensional data for improved disease discrimination.

The adoption of multimodal integration has allowed models to fuse clinical, histopathological, imaging, and molecular data, enhancing predictive power. For instance, Luo-based models combining transcriptomic hub-gene signatures with clinical features reached accuracies exceeding 0.90 and AUCs above 0.94 in treatment response prediction (23). Similarly, Zheng et al. applied convolutional neural networks to digitized biopsy slides, integrating histologic patterns with clinical metadata to achieve high accuracy (> 90%) in glomerular lesion classification (6). In another domain, the radiomics-ultrasound model of Qin et al. combined imaging texture features with serologic markers to estimate LN activity, reporting AUC = 0.95 in training and AUC = 0.77 in testing cohorts (4). These multimodal designs help capture complex interactions across different biological scales.

Recent studies have further expanded this multimodal paradigm by linking molecular discovery with translational applications. Zhang et al. integrated phosphorylation-related gene signatures and single-cell ML analysis to uncover key molecular pathways driving podocyte injury and immune dysregulation in LN, highlighting phospho-signaling networks as novel therapeutic targets (36). In a complementary direction, Mou et al. implemented 12 distinct machine learning algorithms combined with non-negative matrix factorization (NMF) to achieve highly stable transcriptomic-based LN prediction, demonstrating the reproducibility of molecular classifiers across independent datasets (21). Furthermore, the same group developed an integrative framework combining genomics and artificial intelligence to identify mRNA vaccine targets for LN, revealing immune-modulatory peptides capable of rebalancing T- and B-cell signaling (37). Collectively, these findings illustrate how next-generation ML approaches are transcending diagnostic boundaries to enable in silico therapeutic discovery, biomarker-driven immunomodulation, and ultimately, precision nephrology.

The non-invasive potential afforded by machine learning is particularly appealing in LN, where repeated kidney biopsies present risks and are not feasible for longitudinal monitoring. Models employing imaging (e.g., radiomics) or blood-derived features propose alternatives to invasive sampling. For example, Qin et al.’s radiomics-ML system replaces the need for invasive indices by leveraging ultrasound-derived features. In another direction, biomarker-ML frameworks such as Guo et al. propose point-of-care systems that infer renal status from peripheral biomarkers, reducing reliance on biopsy (5). These approaches hold promise for safer, repeatable monitoring in clinical care.

Despite remarkable progress, current machine learning (ML) applications in lupus nephritis (LN) face significant methodological and translational limitations. A major challenge is data scarcity and imbalance, as most studies rely on small, single-center datasets with limited ethnic diversity. This lack of representativeness restricts statistical power and generalizability. For example, Stojanowski et al. trained their neural network on only 58 patients, raising concerns about overfitting and external validity (34). Moreover, algorithms often perform well on internal validation but deteriorate when applied to external or multiethnic cohorts, as differences in disease prevalence, laboratory ranges, and data acquisition methods distort predictive accuracy (34, 38). Ueda et al. emphasized that unbalanced data and underrepresentation of minority populations may amplify disparities, particularly when fairness auditing is not systematically implemented (34).

Another limitation involves model interpretability and transparency. Deep learning architectures, while highly accurate, often function as “black boxes,” offering limited insight into decision-making processes. This opacity undermines clinical trust and hinders regulatory adoption. Recent ethical analyses have stressed that explainability—through attention mapping, SHAP analysis, or transparent reporting of training data—is essential to ensure clinical accountability and reproducibility (38–40). As noted by Hanna et al. and Hoche et al., interpretable ML frameworks are indispensable for safe deployment in healthcare, especially when predictions directly influence treatment selection (38, 39).

Finally, external validation and prospective integration remain largely absent. Few LN models have been tested across multiple institutions or in real-time clinical environments. Without multicentric validation, transportability across platforms and patient populations cannot be ensured. Ratti et al. and Yu et al. argue that ethical deployment requires rigorous cross-validation under diverse clinical conditions and transparent reporting of algorithmic lineage and assumptions (40, 41). Addressing these limitations will require collaborative data sharing, federated learning architectures, and harmonization of reporting standards to foster reliable, equitable, and clinically interpretable ML applications in lupus nephritis.

From an ethical standpoint, issues of privacy, equity, and transparency emerge prominently. The use of patient-level health data invites concerns of re-identification and data misuse unless strong de-identification practices are enforced (38). Algorithmic bias may exacerbate existing health inequalities if models produce systematically worse predictions for underrepresented groups (e.g., by race or socioeconomic status). The emerging literature proposes frameworks for fairness in clinical ML, emphasizing the need to audit for performance disparities across subgroups and mitigate bias through methods such as reweighting or fairness constraints (39, 40). Moreover, AI systems must ensure transparency and accountability: clinicians and patients require interpretability of model decisions and documentation of lineage, training data and assumptions, as mandated by regulatory and ethical frameworks (40, 41). A recent scoping review of AI ethics in healthcare highlights that as clinical AI moves from experimental to real-world applications, previous generalized ethical principles must be reframed to address domain-specific challenges in privacy, consent, accountability, and social risk (42).

Significant gaps in the literature remain. There is a paucity of multicentric and prospective validation of ML models in LN—few studies have tested models in independent cohorts across geographic regions. Longitudinal and time-series modeling is underdeveloped: while some approaches like LSTM models begin to address temporal dynamics, the majority of models remain static snapshots. Few studies integrate federated learning or privacy-preserving methods that allow cross-institutional model training without data sharing. Finally, inclusion of emerging therapies (e.g., CAR-T, RNA vaccines) into ML outcomes is rare in the literature, limiting model relevance for evolving clinical paradigms.

The future application of machine learning (ML) in lupus nephritis (LN) requires transitioning from experimental validation to routine clinical integration. Despite the encouraging predictive performance observed in research settings, the real challenge lies in embedding ML-based systems into clinical trials and daily practice, ensuring interpretability, reproducibility, and regulatory compliance. Oates et al. recently emphasized that predictive models can accelerate adaptive trial designs by identifying high-risk subgroups for early intervention, thereby improving trial efficiency and patient stratification (43). Similarly, the incorporation of ML into clinical decision support systems has shown that algorithms combining longitudinal laboratory and histological data can outperform traditional clinician-based assessments in predicting flares and renal decline (2). However, for these systems to be clinically actionable, their validation must occur under prospective, multicenter conditions with diverse populations reflecting real-world heterogeneity (44).

Another emerging direction involves the implementation of federated learning models and enhanced electronic health record (EHR) interoperability, enabling multi-institutional training without compromising patient privacy. Federated approaches allow distributed data analysis, maintaining data sovereignty and aligning with international data protection standards such as GDPR and HIPAA (45). In 2024, Cheng et al. demonstrated the feasibility of federated frameworks in nephrology by aggregating histopathological data from five hospitals, achieving AUC values above 0.88 for flare prediction while preserving data confidentiality (46). This collaborative paradigm promotes model generalizability and ethical data sharing, overcoming one of the most persistent obstacles in ML-based nephrology research.

Equally important is the longitudinal and multicentric validation of models. Most current studies remain cross-sectional and lack dynamic temporal modeling. Longitudinal validation would allow ML systems to capture disease trajectories and anticipate transitions between quiescent and active states. Recent work applying recurrent neural networks (RNNs) to time-series data of renal biomarkers and treatment courses has achieved promising accuracy in forecasting renal relapse within 6–12 months (20). Expanding these approaches to multicenter settings will be essential to ensure clinical reliability and algorithmic fairness across different demographic and genetic backgrounds.

Finally, future investigations should explore the intersection between ML-guided prediction and novel therapeutic modalities, such as CAR-T and mRNA-based therapies. ML has already been used to identify molecular signatures predictive of therapeutic response, guiding personalized treatment selection (23). In the context of LN, integrating transcriptomic and single-cell RNA sequencing data may enable predictive modeling of immune reconstitution and drug responsiveness. Wu et al. recently developed an ML-integrated framework for mapping cellular pathways affected by CAR-T therapies in autoimmune disease models, providing a translational foundation for nephrology applications (47). Similarly, the emerging use of AI in designing mRNA vaccine targets for SLE and LN offers a glimpse of a new precision-therapeutic paradigm (37).

5 Conclusion

Machine learning (ML) has emerged as a transformative tool for the diagnosis, risk stratification, therapeutic response prediction, and monitoring of lupus nephritis (LN). The models reviewed in this study demonstrate strong discriminative performance, with AUC values frequently exceeding 0.90 across multiple cohorts— particularly in applications targeting non-invasive histological classification, renal flare prediction, and treatment response estimation. These approaches represent a decisive step toward precision nephrology, integrating clinical, histopathological, and molecular information into dynamic and actionable frameworks.

Current evidence suggests that deep learning and multimodal architectures capture the biological and clinical complexity of LN beyond the capabilities of conventional tools. Neural network systems applied to digitized biopsies and radiomic ultrasound models have shown diagnostic performances comparable or superior to traditional methods. Likewise, integrating transcriptomic and serological variables through supervised algorithms provides novel avenues for personalizing therapeutic regimens and anticipating disease relapse, thereby reducing dependence on invasive renal biopsies.

Nevertheless, the clinical translation of ML in LN remains constrained by key structural limitations: the lack of multicenter and longitudinal validation, limited methodological standardization, and underrepresentation of certain demographic groups within training datasets. These deficiencies hinder generalizability and real-world applicability. Furthermore, unresolved ethical and regulatory challenges—including transparency, fairness, and data governance—must be addressed systematically before these models can be safely implemented in patient care.

Future research must prioritize prospective external validation, the deployment of federated and collaborative learning frameworks, and the incorporation of robust ethical and regulatory principles to ensure model accountability and trustworthiness. The convergence of artificial intelligence, digital pathology, and advanced immunotherapies offers a paradigm shift in which lupus nephritis management transitions from reactive treatment to predictive, personalized, and precision-guided care.

Author contributions

DG-B: Formal analysis, Writing – review & editing, Validation, Data curation, Writing – original draft, Investigation. AA-C: Data curation, Writing – review & editing, Formal analysis, Writing – original draft, Investigation, Validation. SA-P: Methodology, Writing – original draft, Data curation, Investigation, Conceptualization, Formal analysis, Writing – review & editing. GA-M: Project administration, Writing – review & editing, Supervision, Conceptualization, Writing – original draft. CM: Writing – review & editing, Supervision, Project administration, Writing – original draft, Conceptualization. RN-Q: Writing – original draft, Conceptualization, Writing – review & editing, Data curation, Formal analysis. AD-V: Formal analysis, Data curation, Methodology, Writing – review & editing, Conceptualization, Writing – original draft. HG-T: Software, Methodology, Supervision, Writing – original draft, Formal analysis, Data curation, Conceptualization, Writing – review & editing.

Funding

The author(s) declare that no financial support was received for the research and/or publication of this article.

Conflict of interest

AD-V and HG-T were employed by Data & Project Consulting Service SAS.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The authors declare that Gen AI was used in the creation of this manuscript. The author(s) verify and take full responsibility for the use of generative AI in the preparation of this manuscript. Generative AI was used: The manuscript was developed with the support of a multi-agent assistant based on large language models (LLMs), designed, validated, and implemented by the corresponding author as part of their doctoral thesis work.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

1. Wang, D-C, Xu, W-D, Wang, S-N, Wang, X, Leng, W, Fu, L, et al. Lupus nephritis or not? A simple and clinically friendly machine learning pipeline to help diagnosis of lupus nephritis. Inflamm Res. (2023) 72:1315–24. doi: 10.1007/s00011-023-01755-7

PubMed Abstract | Crossref Full Text | Google Scholar

2. Chen, Y, Huang, S, Chen, T, Liang, D, Yang, J, Zeng, C, et al. Machine learning for prediction and risk stratification of lupus nephritis renal flare. Am J Nephrol. (2021) 52:152–60. doi: 10.1159/000513566

PubMed Abstract | Crossref Full Text | Google Scholar

3. Guo, H-Q, Wang, X-T, Yang, X, Huang, M-W, and Bai, J. Risk factors for poor outcomes in adult patients with lip through-and-through wounds. Asian J Surg. (2024) 3:133. doi: 10.1016/j.asjsur.2024.11.133

PubMed Abstract | Crossref Full Text | Google Scholar

4. Qin, X, Xia, L, Zhu, C, Hu, X, Xiao, W, Xie, X, et al. Noninvasive evaluation of lupus nephritis activity using a Radiomics machine learning model based on ultrasound. J Inflamm Res. (2023) 16:433–41. doi: 10.2147/JIR.S398399

Crossref Full Text | Google Scholar

5. Guo, J, Teymur, A, Tang, C, Saxena, R, and Wu, T. Advancing point-of-care diagnosis: digitalizing combinatorial biomarker signals for lupus nephritis. Biosensors. (2024) 14:147. doi: 10.3390/bios14030147

PubMed Abstract | Crossref Full Text | Google Scholar

6. Zheng, Z, Zhang, X, Ding, J, Zhang, D, Cui, J, Fu, X, et al. Deep learning-based artificial intelligence system for automatic assessment of glomerular pathological findings in lupus nephritis. Diagnostics. (2021) 11:1983. doi: 10.3390/diagnostics11111983

Crossref Full Text | Google Scholar

7. Austin, HA, Klippel, JH, Balow, JE, le Riche, NG, Steinberg, AD, Plotz, PH, et al. Therapy of lupus nephritis. Controlled trial of prednisone and cytotoxic drugs. N Engl J Med. (1986) 314:614–9. doi: 10.1056/NEJM198603063141004

PubMed Abstract | Crossref Full Text | Google Scholar

8. An, Y, Zhang, H, and Liu, Z. Individualizing therapy in lupus nephritis. Kidney Int Rep. (2019) 4:1366–72. doi: 10.1016/j.ekir.2019.08.005

PubMed Abstract | Crossref Full Text | Google Scholar

9. Chen, LY, Shi, ZR, Tan, GZ, Han, YF, Tang, ZQ, and Wang, L. Systemic lupus erythematosus with and without a family history: a Meta-analysis. Lupus. (2018) 27:716–21. doi: 10.1177/0961203317739133

PubMed Abstract | Crossref Full Text | Google Scholar

10. Barr, RG, Seliger, S, Appel, GB, Zuniga, R, D’Agati, V, Salmon, J, et al. Prognosis in proliferative lupus nephritis: the role of socio-economic status and race/ethnicity. Nephrol Dial Transplant. (2003) 18:2039–46. doi: 10.1093/ndt/gfg345

PubMed Abstract | Crossref Full Text | Google Scholar

11. Stojanowski, J, Konieczny, A, Rydzyńska, K, Kasenberg, I, Mikołajczak, A, Gołębiowski, T, et al. Artificial neural network - an effective tool for predicting the lupus nephritis outcome. BMC Nephrol. (2022) 23:381. doi: 10.1186/s12882-022-02978-2

Crossref Full Text | Google Scholar

12. Zhao, Y, Smith, D, and Jorge, A. Comparing two machine learning approaches in predicting lupus hospitalization using longitudinal data. Sci Rep. (2022) 12:16424. doi: 10.1038/s41598-022-20845-w

Crossref Full Text | Google Scholar

13. Zhang, J, Chen, B, Liu, J, Chai, P, Liu, H, Chen, Y, et al. Predictive modeling of co-infection in lupus nephritis using multiple machine learning algorithms. Sci Rep. (2024) 14:9242. doi: 10.1038/s41598-024-59717-w

Crossref Full Text | Google Scholar

14. Akhgar, A, Sinibaldi, D, Zeng, L, Farris, AB, Cobb, J, Battle, M, et al. Urinary markers differentially associate with kidney inflammatory activity and chronicity measures in patients with lupus nephritis. Lupus Sci Med. (2023) 10:747. doi: 10.1136/lupus-2022-000747

PubMed Abstract | Crossref Full Text | Google Scholar

15. Pesce, F, Pasculli, D, Pasculli, G, De Nicola, L, Cozzolino, M, Granata, A, et al. “The disease awareness innovation network” for chronic kidney disease identification in general practice. J Nephrol. (2022) 35:2057–65. doi: 10.1007/s40620-022-01353-6

PubMed Abstract | Crossref Full Text | Google Scholar

16. Choi, MY, Chen, I, Clarke, AE, Fritzler, MJ, Buhler, KA, Urowitz, M, et al. Machine learning identifies clusters of longitudinal autoantibody profiles predictive of systemic lupus erythematosus disease outcomes. Ann Rheum Dis. (2023) 82:927–36. doi: 10.1136/ard-2022-223808

Crossref Full Text | Google Scholar

17. Zhan, K, Buhler, KA, Chen, IY, Fritzler, MJ, and Choi, MY. Systemic lupus in the era of machine learning medicine. Lupus Sci. Med. (2024) 11:1140. doi: 10.1136/lupus-2023-001140

PubMed Abstract | Crossref Full Text | Google Scholar

18. Abuabara-Franco, E, Bohórquez-Rivero, J, Restom-Arrieta, J, Sáenz-López, J, Gómez-Franco, A, and Navarro-Quiróz, R. Importancia de Un Modelo de Nefroprevención Adaptado Para Colombia. Rev Colomb Nefrol. (2021) 8:e399. doi: 10.22265/acnef.8.3.399

Crossref Full Text | Google Scholar

19. González-Torres, HJ, Yosa Reyes, J, and Montoya Villegas, JC. Implementation of statistical analysis in biomedical sciences through an interactive multi-agent assistant based on large language models / Implementación Del Análisis Estadístico En Ciencias Biomédicas Mediante un Asistente Interactivo Multiagente Basado. Cali, Colombia: Universidad del Valle (2025).

Google Scholar

20. Huang, S, Chen, Y, Song, Y, Wu, K, Chen, T, Zhang, Y, et al. Deep learning model to predict lupus nephritis renal flare based on dynamic multivariable time-series data. BMJ Open. (2024) 14:e071821. doi: 10.1136/bmjopen-2023-071821

PubMed Abstract | Crossref Full Text | Google Scholar

21. Mou, L, Lu, Y, Wu, Z, Pu, Z, Huang, X, and Wang, M. Applying 12 machine learning algorithms and non-negative matrix factorization for robust prediction of lupus nephritis. Front Immunol. (2024) 15:1218. doi: 10.3389/fimmu.2024.1391218

PubMed Abstract | Crossref Full Text | Google Scholar

22. Tang, Y, Zhang, W, Zhu, M, Zheng, L, Xie, L, Yao, Z, et al. Lupus nephritis pathology prediction with clinical indices. Sci Rep. (2018) 8:10231. doi: 10.1038/s41598-018-28611-7

Crossref Full Text | Google Scholar

23. Lee, D-J, Tsai, P-H, Chen, C-C, and Dai, Y-H. Incorporating knowledge of disease-defining hub genes and regulatory network into a machine learning-based model for predicting treatment response in lupus nephritis after the first renal flare. J Transl Med. (2023) 21:76. doi: 10.1186/s12967-023-03931-z

Crossref Full Text | Google Scholar

24. Wang, Z, Hu, D, Pei, G, Zeng, R, and Yao, Y. Identification of driver genes in lupus nephritis based on comprehensive bioinformatics and machine learning. Front Immunol. (2023) 14:1288699. doi: 10.3389/fimmu.2023.1288699

Crossref Full Text | Google Scholar

25. Tang, C, Tan, G, Teymur, A, Guo, J, Haces-Garcia, A, Zhu, W, et al. A serum biomarker panel and Miniarray detection system for tracking disease activity and flare risk in lupus nephritis. Front Immunol. (2025) 16:1907. doi: 10.3389/fimmu.2025.1541907

PubMed Abstract | Crossref Full Text | Google Scholar

26. Deng, Y, Pacheco, JA, Ghosh, A, Chung, A, Mao, C, Smith, JC, et al. Natural language processing to identify lupus nephritis phenotype in electronic health records. BMC Med Inform Decis Mak. (2024) 22:348. doi: 10.1186/s12911-024-02420-7

Crossref Full Text | Google Scholar

27. Li, T, Lin, S, Guan, Z, Zhou, Y, Zeng, D, Wang, Z, et al. A deep learning system for detecting systemic lupus erythematosus from retinal images. Cell Reports Med. (2025) 6:102203. doi: 10.1016/j.xcrm.2025.102203

PubMed Abstract | Crossref Full Text | Google Scholar

28. Izadi, Z, Gianfrancesco, M, Anastasiou, C, Schmajuk, G, and Yazdany, J. Development and validation of a risk scoring system to identify patients with lupus nephritis in electronic health record data. Lupus Sci Med. (2024) 11:e001170. doi: 10.1136/lupus-2024-001170

PubMed Abstract | Crossref Full Text | Google Scholar

29. Navarro Quiroz, E, Chavez-Estrada, V, Macias-Ochoa, K, Ayala-Navarro, MF, Flores-Aguilar, AS, Morales-Navarrete, F, et al. Epigenetic mechanisms and posttranslational modifications in systemic lupus erythematosus. Int J Mol Sci. (2019) 20:5679. doi: 10.3390/ijms20225679

Crossref Full Text | Google Scholar

30. Navarro-Quiroz, E, Pacheco-Lugo, L, Lorenzi, H, Díaz-Olmos, Y, Almendrales, L, Rico, E, et al. High-throughput sequencing reveals circulating MiRNAs as potential biomarkers of kidney damage in patients with systemic lupus erythematosus. PLoS One. (2016) 11:e0166202. doi: 10.1371/journal.pone.0166202

PubMed Abstract | Crossref Full Text | Google Scholar

31. Gomathi, S., and Narayani, V. Implementing big data analytics to predict systemic lupus erythematosus. In Proceedings of the 2015 international conference on innovations in information, embedded and communication systems (ICIIECS); India: IEEE, (2015); pp. 1–5.

Google Scholar

32. Agrawal, D., Das, S., and El Abbadi, A. Big data and cloud computing. In Proceedings of the proceedings of the 14th international conference on extending database technology; ACM: New York, NY, USA, (2011); pp. 530–533.

Google Scholar

33. Petri, M, Orbai, A-M, Alarcón, GS, Gordon, C, Merrill, JT, Fortin, PR, et al. Derivation and validation of the systemic lupus international collaborating clinics classification criteria for systemic lupus erythematosus. Arthritis Rheum. (2012) 64:2677–86. doi: 10.1002/art.34473

PubMed Abstract | Crossref Full Text | Google Scholar

34. Ueda, D, Kakinuma, T, Fujita, S, Kamagata, K, Fushimi, Y, Ito, R, et al. Fairness of artificial intelligence in healthcare: review and recommendations. Jpn J Radiol. (2024) 42:3–15. doi: 10.1007/s11604-023-01474-3

PubMed Abstract | Crossref Full Text | Google Scholar

35. Yang, P, Liu, Z, Lu, F, Sha, Y, Li, P, Zheng, Q, et al. Machine learning models predicts risk of proliferative lupus nephritis. Front Immunol. (2024) 15:1413569. doi: 10.3389/fimmu.2024.1413569

PubMed Abstract | Crossref Full Text | Google Scholar

36. Mou, L, Chen, Z, Tian, X, Lai, Y, Pu, Z, and Wang, M. Phosphorylation-related genes in lupus nephritis: single-cell and machine learning insights. Genes Dis. (2025) 12:101385. doi: 10.1016/j.gendis.2024.101385

PubMed Abstract | Crossref Full Text | Google Scholar

37. Mou, L, Lu, Y, Wu, Z, Pu, Z, and Wang, M. Integrating genomics and AI to uncover molecular targets for MRNA vaccine development in lupus nephritis. Front Immunol. (2024) 15:1445. doi: 10.3389/fimmu.2024.1381445

PubMed Abstract | Crossref Full Text | Google Scholar

38. Hanna, MG, Pantanowitz, L, Jackson, B, Palmer, O, Visweswaran, S, Pantanowitz, J, et al. Ethical and Bias considerations in artificial intelligence/machine learning. Mod Pathol. (2025) 38:100686. doi: 10.1016/j.modpat.2024.100686

PubMed Abstract | Crossref Full Text | Google Scholar

39. Hoche, M, Mineeva, O, Rätsch, G, Vayena, E, and Blasimme, A. What makes clinical machine learning fair? A practical ethics framework. PLOS Digit Health. (2025) 4:e0000728. doi: 10.1371/journal.pdig.0000728

PubMed Abstract | Crossref Full Text | Google Scholar

40. Ratti, E, Morrison, M, and Jakab, I. Ethical and social considerations of applying artificial intelligence in healthcare—a two-pronged scoping review. BMC Med Ethics. (2025) 26:68. doi: 10.1186/s12910-025-01198-1

Crossref Full Text | Google Scholar

41. Yu, S, Lee, S-S, and Hwang, H. The ethics of using artificial intelligence in medical research. Kosin Med J. (2024) 39:229–37. doi: 10.7180/kmj.24.140

Crossref Full Text | Google Scholar

42. Abujaber, AA, and Nashwan, AJ. Ethical framework for artificial intelligence in healthcare research: a path to integrity. World J Methodol. (2024) 14:94071. doi: 10.5662/wjm.v14.i3.94071

PubMed Abstract | Crossref Full Text | Google Scholar

43. Wolf, BJ, Spainhour, JC, Arthur, JM, Janech, MG, Petri, M, and Oates, JC. Development of biomarker models to predict outcomes in lupus nephritis. Arthritis Rheumatol. (2016) 68:1955–63. doi: 10.1002/art.39623

PubMed Abstract | Crossref Full Text | Google Scholar

44. Rojas-Rivera, JE, García-Carro, C, Ávila, AI, Espino, M, Espinosa, M, Fernández-Juárez, G, et al. Diagnosis and treatment of lupus nephritis: a summary of the consensus document of the Spanish Group for the Study of glomerular diseases (GLOSEN). Clin Kidney J. (2023) 16:1384–402. doi: 10.1093/ckj/sfad055

PubMed Abstract | Crossref Full Text | Google Scholar

45. Brisimi, TS, Chen, R, Mela, T, Olshevsky, A, Paschalidis, IC, and Shi, W. Federated learning of predictive models from federated electronic health records. Int J Med Inform. (2018) 112:59–67. doi: 10.1016/j.ijmedinf.2018.01.007

PubMed Abstract | Crossref Full Text | Google Scholar

46. Cheng, C, Li, B, Li, J, Wang, Y, Xiao, H, Lian, X, et al. Multi-stain deep learning prediction model of treatment response in lupus nephritis based on renal histopathology. Kidney Int. (2025) 107:714–27. doi: 10.1016/j.kint.2024.12.007

PubMed Abstract | Crossref Full Text | Google Scholar

47. Wu, D, Xu-Monette, ZY, Zhou, J, Yang, K, Wang, X, Fan, Y, et al. CAR T-cell therapy in autoimmune diseases: a promising frontier on the horizon. Front Immunol. (2025) 16:878. doi: 10.3389/fimmu.2025.1613878

PubMed Abstract | Crossref Full Text | Google Scholar

48. Yin, C, Xiao, W, Hu, X, Liu, X, Xian, H, Su, J, et al. Non-invasive prediction of the chronic degree of lupus nephropathy based on ultrasound radiomics. Lupus. (2024) 33:121–128. doi: 10.1177/09612033231223373

PubMed Abstract | Crossref Full Text | Google Scholar