Abstract Abstract Objective To investigate the causal relationship between serotonin levels, adenosine deaminase (ADA) activity and multiple sclerosis (MS) progression using an integrative multi-omics approach. Methods and analysis A two-sample Mendelian randomisation (MR) analysis was performed using inverse variance weighted (IVW) estimation to assess causality between serotonin, ADA and MS risk. Single-cell transcriptomic data from the Gene Expression Omnibus ([30]GSE194078) were analysed to identify ADA-expressing immune cell subpopulations. Moreover, machine learning algorithms (Support Vector Machine-Recursive Feature Elimination, Least Absolute Shrinkage and Selection Operator and random forest) were applied to identify diagnostic biomarkers, following which a nomogram was constructed and validated. Results MR analysis revealed that serotonin levels were positively correlated with MS progression (IVW β=0.350, p=3.63E-05), whereas genetically predicted ADA levels were inversely associated with MS risk (IVW β=−0.395, p=2.73E-04). Additionally, serotonin levels exhibited an inverse causal relationship with ADA activity (IVW β=−0.089, p=8.70E-03), with no evidence of reverse causation. Single-cell analysis identified 18 cellular subpopulations and six major immune cell types, with ADA highly expressed in T-NK cells and expressed at lower levels in platelets. Meanwhile, ADA expression was higher in the low immune receptor signalling group. Enrichment analysis indicated that differentially expressed genes were enriched in biological processes such as cytoplasmic translation and RNA splicing, as well as Kyoto Encyclopedia of Genes and Genome pathways such as Ribosome and Neurodegeneration-Multiple Diseases. Three key feature genes (IK, UBA52 and CCDC25) were identified, and the nomogram based on these genes demonstrated high diagnostic accuracy, with an AUC of 1.000 in the training dataset and 0.976 in the validation dataset. Conclusions Serotonin promotes MS progression by inhibiting ADA activity, positioning the serotonin-ADA axis as a potential therapeutic target. The identified biomarkers (IK, UBA52 and CCDC25) and the constructed nomogram may enhance diagnostic precision for MS, providing valuable insights for MS management and laying a theoretical reference for future studies. Keywords: Multiple sclerosis, Chronic Disease, Pathology, GENETICS __________________________________________________________________ STRENGTHS AND LIMITATIONS OF THIS STUDY. * This study integrated multi-omics approaches (Mendelian randomisation, single-cell transcriptomics and machine learning) to systematically characterise the serotonin-adenosine deaminase axis in multiple sclerosis. * Inverse variance weighted estimation and comprehensive sensitivity analyses ensured robust causal inference, with machine learning identifying core biomarkers. * The predominant reliance on European-derived genomic data may introduce population stratification bias, thereby limiting the generalisability of the findings to diverse ethnic groups. * While single-cell sequencing provided high-resolution insights into cellular heterogeneity, technical artefacts associated with high-throughput omics and Mendelian randomisation assumptions are methodological constraints. * The findings require validation in large-scale, prospective clinical studies to corroborate translational relevance. Introduction As is well documented, multiple sclerosis (MS) is a chronic and heterogeneous autoimmune disorder hallmarked by immune-mediated demyelination and neurodegeneration within the central nervous system (CNS),[31]^1 2 resulting in a broad spectrum of neurological symptoms and varying degrees of disability.[32]^3 4 Its pathogenesis involves genetic susceptibility, environmental factors (such as EB virus infection, vitamin D deficiency, smoking and high sodium diets) and abnormal immune responses.[33]^5 Immunologically, autoreactive CD4 T cells attack myelin, with Th1 and Th17 cells playing crucial roles in triggering neuroinflammation and Treg cells exerting immunosuppressive effects.[34]^6 B cells contribute to the pathological process through both antibody-dependent and antibody-independent mechanisms, promoting the production of autoantibodies and activating T cells, which further exacerbate the inflammatory response. Prolonged microglial activation results in sustained inflammation and damage to neural tissue.[35]^7 Recently, the presence of exhausted T cells, characterised by reduced effector function and persistent expression of inhibitory receptors, has been observed in MS lesions, suggesting a potential role in chronic inflammation and immune dysregulation.[36]^8 Although current therapies such as interferon-β and Fingolimod can alleviate symptoms,[37]^9 10 side effects and inter-individual variability in treatment responses highlight the pressing need for the development of novel therapeutic targets and strategies.[38]^11 Serotonin, a neurotransmitter that participates in mood regulation, also exerts immunomodulatory effects.[39]^12 13 Meanwhile, the dysregulation of adenosine deaminase (ADA), an enzyme involved in purine metabolism, has been associated with various immunological disorders.[40]^14 15 Thus, the interaction between serotonin levels and ADA activity presents a novel axis with potential therapeutic implications, warranting further mechanistic explorations and clinical investigation in the context of MS. At present, cutting-edge methodologies such as single-cell sequencing,[41]^16 Mendelian randomisation (MR) analysis[42]^17 and advanced machine-learning models hold significant promise in elucidating the intricate relationship between serotonin levels, ADA inhibition and MS progression.[43]^18 Indeed, MR has been widely applied in biomedical studies[44]^19 20 and provides a robust framework for inferring causal relationships between genetic variants, serotonin levels, ADA activity and MS risk.[45]^21 Furthermore, the application of sophisticated machine-learning models enables the integration and analysis of complex datasets, revealing intricate biological patterns and interactions underlying disease mechanisms.[46]^22 Taken together, these advanced methodologies offer a comprehensive approach to unravelling the role of the serotonin–ADA axis in MS, laying a reference for the development of novel therapeutic strategies and personalised medicine approaches in the management of this debilitating condition. Materials and methods Research and design scheme In this study, a two-sample MR analysis was performed to elucidate the causal relationships between 91 inflammatory cytokines, 1400 circulating plasma metabolites and MS. Furthermore, the MR analysis was expanded to specifically assess causal links between metabolite levels and MS, thus enabling an evaluation of their potential influence on the observed associations. This approach facilitated an in-depth examination of the intricate relationship between metabolite levels and MS, mediated through the regulation of inflammatory cytokine levels. The validity of MR-based causal inference is predicated on three foundational assumptions: a robust association between genetic variants and exposure variables,[47]^23 the absence of potential confounders that influence the relationship between genetic variants and the outcome (ie, MS) and the critical premise that the exposure (metabolite levels) represents the sole pathway through which genetic variants affect the outcome, thereby minimising pleiotropic effects.[48]^24 25 This rigorous approach strengthened the reliability of causal inferences, which were subsequently corroborated at the single-cell level, thereby elucidating the intricate interactions between metabolite levels and MS progression through the modulation of inflammatory cytokines ([49]figure 1). Figure 1. Study flowchart. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; LASSO, Least Absolute Shrinkage and Selection Operator; PCA, principal component analysis; SVM-RFE, Support Vector Machine-Recursive Feature Elimination; UMAP, Uniform Manifold Approximation and Projection. [50]Figure 1 [51]Open in a new tab Sources of exposure and outcome data Exposure data for inflammatory cytokines were derived from a genome-wide protein quantitative trait loci (pQTL) analysis of 91 plasma proteins measured in 14 824 participants using the Olink Target platform.[52]^26 Data on 1091 distinct metabolites and 309 metabolite ratios were retrieved from extensive genome-wide association studies (GWAS) involving 8299 individuals in the Canadian Longitudinal Study on Ageing cohort.[53]^27 Both cytokine and metabolite data included summary statistics (ie, effect sizes and P values) for genetic variants associated with their levels. For MS outcomes, GWAS summary statistics (including ORs and p values) for genetic variants linked to MS risk were sourced from the International MS Genetics Consortium, covering 47 429 cases and 68 374 controls of European ancestry.[54]^28 All datasets analysed in this study were retrieved from publicly accessible GWAS databases; given their open-access and de-identified nature, the requirement for ethical approval was waived. In the present study, MS progression was indirectly evaluated using cross-sectional GWAS data on MS risk (case-control status), given that longitudinal data on disease progression were not available in the public datasets. While MS risk and progression are related, they represent distinct aspects of the disease. The findings of this study may provide valuable insights into the genetic determinants of MS susceptibility, which may inform mechanisms underlying disease progression. Serotonin levels were estimated using genetic variants associated with serum serotonin metabolites. ADA activity was inferred from genetic variants linked to ADA enzyme activity or gene expression levels in blood samples, as reported in prior GWAS and pQTL studies. Collection and analysis of single-cell transcriptome sequencing data The single-cell transcriptomic dataset for MS was obtained from the Gene Expression Omnibus (GEO)[55]^29 database, specifically the [56]GSE194078 dataset. Data extraction was conducted using the R programming language in conjunction with the Seurat software package.[57]^30 Quality control parameters were established as follows: (1) exclusion of cells expressing fewer than 200 or over 4000 genes and (2) exclusion of cells with mitochondrial gene expression accounting for over 10% of total gene expression. Following preprocessing, gene expression data were normalised. Seurat was employed for quality control of the single-cell dataset. Initially, principal component analysis (PCA) was employed for dimensionality reduction, followed by further reduction using the Uniform Manifold Approximation and Projection (UMAP) algorithm. Based on the results of PCA,[58]^31 32 single-cell clustering was visualised via UMAP,[59]^33 and subpopulation cell clustering was further refined using t-Distributed Stochastic Neighbour Embedding (t-SNE).[60]^34 The MonacoImmuneData tool was used to annotate cellular subpopulations. The subsequent analysis focused on the levels of metabolic receptor signalling (encompassing transport and uptake) and alterations in inflammatory factor expression. Immune cell populations were stratified into two cohorts based on expression levels: one with elevated receptor signal expression and the other with low expression. Stringent selection criteria (p<0.05, Log2 Fold Change >0.25) were applied to identify differentially expressed genes (DEGs) in these cohorts. Functional enrichment analysis Functional enrichment analysis was performed to systematically characterise DEGs and identify potential functional targets, as described in previous studies.[61]35,[62]37 Gene Ontology (GO) analysis was performed to annotate genes across three primary domains, namely, molecular function, biological process (BP) and cellular component. Additionally, Kyoto Encyclopedia of Genes and Genome (KEGG) pathway enrichment analysis was carried out to identify biological pathways linked with the DEGs. Selection of signature genes A systematic analysis of DEGs was performed to identify signature genes associated with MS. Several machine-learning methods, including Random Forest[63]^38 and Support Vector Machine-Recursive Feature Elimination (SVM-RFE),[64]^39 which are commonly used for key factor selection, were applied. In addition, Least Absolute Shrinkage and Selection Operator (LASSO) regression was employed to construct linear models and identify significant variables,[65]^40 as outlined in previous studies.[66]^41 42 Finally, a Venn diagram was generated to identify signature genes. Construction and validation of the diagnostic model Based on the identified signature genes, a diagnostic nomogram was constructed to predict the risk of MS. The predictive accuracy of the model was evaluated using calibration curves and decision curve analysis (DCA). Finally, ROC curves were plotted using R software to further assess model performance. The nomogram was trained on feature gene expression data from the training set and validated using an independent dataset with no overlapping samples, ensuring an unbiased evaluation of its predictive performance. Construction of interaction networks for shared feature genes GeneMANIA ([67]http://www.genemania.org/) was used to construct interaction networks among three shared feature genes. The detailed results of these interaction networks are provided in the [68]online supplemental materials. Statistical analysis MR analysis was conducted using the MR software package (version 0.4.3), employing the inverse variance weighted (IVW) method[69]^43 and the weighted median method.[70]^44 The former was employed to estimate the impact of exposure variables on outcomes based on the validity of the MR assumption. Cochran’s Q test was used to assess residual heterogeneity in the IVW model (p<0.05). Additionally, the MR–Egger intercept test was employed to evaluate potential pleiotropy in causal estimates (p<0.05). Data visualisation encompassed scatter plots, funnel plots and forest plots. Scatter plots were generated to demonstrate the consistency of results and the influence of potential outliers. Funnel plots were constructed to assess the robustness of correlations and verify the absence of heterogeneity. Lastly, forest plots provided a visual representation of the interaction between instrumental variables (single-nucleotide polymorphism (SNP)) and study outcomes. In MR analyses, the beta (β) coefficient represents the expected change in the log-odds ratio of MS risk per SD increase in the exposure (serotonin levels or ADA activity). Instrumental variables (SNPs) for serotonin and ADA were selected based on genome-wide significance (p<5×10^−⁸) in GWAS datasets. SNPs were excluded if evidence of horizontal pleiotropy was detected (MR–Egger intercept test, p<0.05) or weak instrument bias was present (F-statistic<10). Patient and public involvement Patients and the public were not involved in the design, conduct, reporting or dissemination of this study. The research used publicly available GWAS datasets and single-cell transcriptomic data from the GEO database, which do not involve direct patient or public participation. No primary data were collected from patients, and the study objectives and outcomes were determined solely by the research team. Results MR analysis of serotonin levels regulating MS progression via inhibition of ADA activity To explore the causal relationship between serotonin-mediated ADA inhibition and MS, a two-sample MR analysis was performed using the IVW method. After selecting and harmonising IVs, eight SNPs were retained for serotonin following the exclusion of palindromic/ambiguous variants, those lacking instrumental utility and SNPs with inconsistent causal directionality ([71]onlinesupplemental tables S1 [72]S2[73]undefined). For ADA, six SNPs were selected ([74]onlinesupplemental tables S3 [75]S4[76]undefined). All retained SNPs exhibited F-statistics exceeding 10, confirming robust instrumental validity. MR results indicated that serotonin promotes MS progression, with statistical significance maintained after false discovery rate (FDR) correction (IVW method:β=0.350, p=3.63E-05, P(FDR)=0.049; [77]online supplemental table S5). In contrast, genetically predicted ADA levels exhibited an inverse association with MS risk, which also remained significant post-FDR adjustment (IVW method: β=−0.395, p=2.73E-04, P(FDR)=0.012; [78]online supplemental table S5). Concurrently, a negative causal relationship was identified between serotonin levels and ADA activity (IVW method: β=−0.089, p=8.70E-03; [79]figure 2A, [80]online supplemental tables S5), with no evidence of reverse causation detected between (IVW method: β=−0.011, p=0.742; [81]online supplemental tables S5). As expected, sensitivity analyses yielded consistent results, with no evidence of pleiotropy-related bias ([82]online supplemental figure S1). The reliability of the findings was further validated using scatter plots ([83]figure 2B–D). Detailed analyses ([84]onlinesupplemental tables S6[85]S9) corroborated the absence of pleiotropic effects. At the same time, Cochran’s Q test revealed the absence of significant heterogeneity ([86]onlinesupplemental tables S10[87]S13). Collectively, these findings suggest that serotonin promotes MS progression by inhibiting ADA activity, supporting the hypothesis of an interaction between serotonin and ADA in the pathogenesis of MS and providing a theoretical foundation for identifying potential therapeutic targets. Figure 2. Scatter plots illustrating the correlations between serotonin levels, ADA activity and MS progression. (A) Association between ADA activity and MS. (B) Relationship between serotonin levels and MS. (C) Correlation between serotonin levels and ADA activity. (D) Forest plot depicting the causal relationship between serotonin levels and ADA activity. ADA, adenosine deaminase; MR, Mendelian randomisation; MS, multiple sclerosis.; SNP, single-nucleotide polymorphism. [88]Figure 2 [89]Open in a new tab Transcriptomic data reveal divergent cellular constituents in MS Single-cell transcriptomic analysis, following dimensionality reduction and clustering, yielded 18 distinct cellular subpopulations ([90]figure 3A) and six major immune cell types, namely, T cells, monocytes, NK cells, B cells, common myeloid progenitors and platelets ([91]figure 3B,C). Next, ADA expression levels were quantified and visualised across these cell types ([92]figure 3D), with the highest levels observed in T-NK cells and the lowest in platelets. Consequently, the ensuing analysis focused on T-NK cells to explore the involvement of immune-related metabolic receptor signalling pathways. Immune cells were stratified into two groups according to receptor signalling activity, namely, the high and low expression groups. Interestingly, ADA expression was elevated in the low receptor signalling group ([93]figure 3E,F). Overall, these results suggest that serotonin may modulate MS progression by regulating ADA expression in coordination with immune receptor signalling pathways. Figure 3. Visualisation of transcriptomic data. (A) UMAP plot displaying 18 clusters. (B) Cell type proportions. (C) ADA levels across different cell types. (D) Distribution of intracellular ADA levels. (E) Proportions of cells with high and low ADA levels in receptor signalling groups. (F) Expression levels of ADA in high and low receptor signalling groups; ADA, adenosine deaminase; UMAP, Uniform Manifold Approximation and Projection. [94]Figure 3 [95]Open in a new tab Enrichment analysis of DEGs Enrichment analysis revealed that the DEGs were primarily enriched in BPs related to cytoplasmic translation, RNA splicing, ATP synthesis coupled with electron transport and protein–RNA complex organisation ([96]figure 4A,B). The most enriched KEGG pathways included ribosome, amyotrophic lateral sclerosis, pathways of neurodegeneration-multiple diseases and chemical carcinogenesis-reactive oxygen species ([97]figure 4C,D). These findings offer insights into the molecular mechanisms underlying related diseases and may assist in the identification of potential biomarkers or therapeutic targets, contributing to the development of future treatment strategies. Figure 4. Functional enrichment analysis of differentially expressed genes. (A)Gene Ontology (GO) bar chart. (B) GO bubble chart. (C) Kyoto Encyclopedia of Genes and Genome (KEGG) bar chart. (D) KEGG bubble chart. [98]Figure 4 [99]Open in a new tab Selection of feature genes Initially, 16 key feature variables were identified using SVM-RFE ([100]figure 5A,B). Logistic regression analysis subsequently identified 15 related genes with statistically significant associations ([101]figure 5C,D). A random forest model, integrated with feature selection techniques, was then applied to evaluate classification error, resulting in the identification of 30 relatively important genes ([102]figure 5E,F). Thereafter, a Venn diagram was plotted to visualise overlaps among the gene sets acquired from these methods ([103]figure 5G), ultimately revealing three shared feature genes, namely, IK, UBA52 and CCDC25. Figure 5. Workflow for Identifying multiple sclerosis-related feature genes; (A, B) Selection and validation of biomarker genes using the Support Vector Machine-Recursive Feature Elimination method. (C) Feature selection using the Least Absolute Shrinkage and Selection Operator regression model. (D) Optimisation of feature selection under minimum absolute shrinkage criteria. (E) Relationship between the number of classification trees and error rates in the random forest model. (F) Relevant feature genes with Millennium Development Goal scores. (H) Screening of feature genes using Venn diagrams. MS, multiple sclerosis. [104]Figure 5 [105]Open in a new tab Construction of an MS diagnostic model based on feature genes Furthermore, a diagnostic nomogram for MS was constructed based on the feature genes IK, UBA52 and CCDC25. The nomogram was visualised ([106]figure 6A), and its performance was evaluated using a calibration curve ([107]figure 6B), and DCA ([108]figure 6C) in R software. The calibration curve illustrated a strong agreement between predicted and observed outcomes, with the ‘Number high risk’ curve closely overlapping the ‘Number high risk with event’ curve. These findings collectively signal that the nomogram exhibits robust predictive performance and high accuracy for MS diagnosis ([109]figure 6D). Figure 6. Construction and validation of the diagnostic model for MS. (A) Nomogram for MS. (B) Calibration curve of the model. (C) Decision curve analysis. (D) Clinical impact curve; MS, multiple sclerosis. [110]Figure 6 [111]Open in a new tab Evaluation of the nomogram using training and validation datasets The diagnostic performance of the nomogram constructed from the three shared feature genes (IK, UBA52 and CCDC25) was evaluated using ROC curve analysis. In the training dataset ([112]GSE13551), the individual AUCs for IK, UBA52 and CCDC25 were 0.985, 0.969 and 0.988, respectively ([113]figure 7A). Notably, the combined nomogram achieved perfect classification performance, with an AUC of 1.000 (95% CI: 1.000 to 1.000) ([114]figure 7B). In the validation dataset ([115]GSE21942), the corresponding AUCs for the three genes were 0.798, 0.867 and 0.819 ([116]figure 7C), with the nomogram maintaining high accuracy with an AUC of 0.976 (95% CI: 0.914 to 1.000) ([117]figure 7D). These results conjointly demonstrate that IK, UBA52 and CCDC25, when integrated into the nomogram, may serve as effective biomarkers for the diagnosis of MS. Figure 7. Validation of the nomogram using receiver operating characteristic (ROC) curves: (A) ROC curve and area under the curve (AUC) for each common feature gene in [118]GSE13551. (B) ROC curve and AUC for the nomogram in [119]GSE13551. (C) ROC curve and AUC for each common feature gene in [120]GSE21942. (D) ROC curve and AUC for the nomogram in [121]GSE21942. [122]Figure 7 [123]Open in a new tab Interaction network of shared feature genes and their co-expressed genes Subsequently, a comprehensive analysis was carried out to elucidate the interaction network and functional associations of the shared feature genes. The results revealed a complex interaction network, with physical interactions comprising the largest proportion (77.64%). Additional interaction types included co-expression (8.01%), predicted interactions (5.37%), co-localisation (3.63%), genetic interactions (2.87%), pathway-related associations (1.88%) and shared protein domain interactions (0.60%). Functionally, the network was primarily involved in processes such as protein targeting to the endoplasmic reticulum, establishment of protein localisation to the endoplasmic reticulum, protein-membrane targeting, cotranslational protein targeting to membranes and cytosolic ribosome activity ([124]figure 8 and Supplementary Materials S1). Figure 8. Analysis of shared feature genes and their co-expression using GeneMANIA. [125]Figure 8 [126]Open in a new tab Discussion This comprehensive analysis highlights the intricate interplay between serotonin levels, ADA activity and MS progression. By integrating MR analysis, single-cell transcriptomic profiling and validation through advanced machine-learning models, this study provides novel insights into the molecular mechanisms underlying MS pathogenesis and the identification of potential therapeutic targets. MR analysis revealed a dual role for serotonin and ADA in MS progression. Specifically, a positive causal relationship was observed between serotonin levels and MS risk, suggesting that elevated serotonin may increase MS susceptibility. In contrast, genetically predicted ADA levels were negatively correlated with MS risk, indicating a potential protective effect against MS susceptibility. Of note, the inverse causal relationship between serotonin and ADA supports the hypothesis that serotonin may partially contribute to MS susceptibility by inhibiting ADA activity. These findings emphasise the potential of targeting the serotonin–ADA axis as a therapeutic strategy for MS. Serotonin,[127]^45 traditionally recognised for its role in mood regulation,[128]^46 also exerts immunomodulatory effects, thereby impacting susceptibility to autoimmune disorders such as MS through a complex array of mechanisms.[129]^47 By interacting with a broad spectrum of 5-hydroxytryptamine (5-HT) receptors across various immune cell types, serotonin can either drive pro-inflammatory cytokine release or exert anti-inflammatory effects, thereby influencing inflammatory processes central to MS pathogenesis.[130]48,[131]50 Additionally, its capacity to alter the permeability of the blood–brain barrier (BBB) facilitates immune cell infiltration into the CNS, further contributing to demyelination and neuronal damage.[132]^12 51 52 Within the CNS, serotonin exerts neuroimmunomodulatory effects and affects microglial and astrocyte activity, thereby fostering a neuroinflammatory environment.[133]^53 Molecularly, its impact is mediated via diverse signalling pathways activated by its receptors, including cyclic adenosine monophosphate (cAMP), phosphoinositide and mitogen-activated protein kinase (MAPK) pathways, which collectively modulate gene expression, cytokine production and cell survival. This intricate interplay of immunological, neuroinflammatory and molecular mechanisms highlights its significant yet complex role in MS progression and its potential as a therapeutic target. A key aspect of serotonin’s action involves its interaction with ADA, where elevated serotonin levels suppress ADA activity, leading to adenosine dysregulation and potentially aggravating inflammatory responses.[134]^54 55 The relationship between ADA and MS encompasses several mechanisms, including the immunomodulatory role of adenosine, which can influence immune cell behaviour and cytokine profiles, thereby affecting MS pathogenesis.[135]^56 57 Dysregulated ADA activity contributes to neuroinflammation within the CNS, exacerbating demyelination and neuronal damage, which are hallmarks of MS.[136]^58 Furthermore, the effect of ADA on BBB integrity and lymphocyte function demonstrates its role in immune cell infiltration into the CNS and the autoimmune response characteristic of MS.[137]^15 By affecting adenosine levels, ADA indirectly mediates glial cell activation and the inflammatory cascade in the CNS, emphasising its multifaceted involvement in the pathophysiology of MS. Single-cell transcriptomic analysis provided additional mechanistic insights by identifying cellular subpopulations and signalling pathways implicated in MS pathogenesis.[138]^31 59 Besides, single-cell transcriptomic data offer profound insights into the characterisation of cell subpopulations and signalling pathways; however, some limitations, such as technical and batch effects, merit acknowledgement.[139]^60 61