{"paper_id":"cc36f256-dffd-4f5e-8e75-4ca3135c2eb8","body_text":"1\nTitle: 1 \nTranslation of miRNA blood-based discovery to molecular testing for clinical diagnosis of 2 \nendometriosis 3 \n 4 \nYanqin Yu1,†, Wing Hing Wong2,†, Yao Hu1, Yirong Shen1, Xiaochun Xu1, Shen Lu1, Libo Zhu3, 5 \nXinmei Zhang3,*, and Farideh Z Bischoff2,*  6 \n 7 \n1Heranova Lifesciences, Hangzhou, China 8 \n2Heranova Lifesciences, Burlington, MA 01803, USA 9 \n3Women’s Hospital, School of Medicine, Zhejiang University, Hangzhou, China 10 \n 11 \n†Equal contributions 12 \n 13 \n*Corresponding Authors 14 \nFarideh Bischoff, PhD 15 \nChief Medical Officer 16 \nHeranova Lifesciences, Inc 17 \nfarideh.bischoff@heranova.com 18 \n 19 \nXinmei Zhang, MD 20 \nProfessor; Chief Physician 21 \nWomen’s Hospital, School of Medicine, Zhejiang University, Hangzhou, China 22 \nzhangxinm@zju.edu.cn   23 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\n 2\nAbstract 24 \nEndometriosis is a common yet often underdiagnosed condition, partly due to the lack of reliable 25 \ndiagnostics. This study examines the clinical feasibility of a blood-based, miRNA-driven test to 26 \ndiagnose endometriosis and address the challenges of translating next-generation sequencing (NGS) 27 \nfindings into clinical use. Serum from 20 patients and 20 controls underwent miRNA sequencing to 28 \nidentify diagnostic biomarkers. A machine learning model built on all NGS-based differentially 29 \nexpressed miRNA biomarkers achieved ≥ 90% accuracy. Validation by qPCR confirmed some but not 30 \nall findings, underscoring the difficulty of adapting NGS discoveries for routine diagnostics. 31 \nNonetheless, serum miRNA biomarkers show strong promise for non-invasive endometriosis 32 \ndetection, with further optimization needed for clinical translation.  33 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 3\nIntroduction 34 \nEndometriosis is a chronic and often debilitating condition that affects approximately 5-10% of 35 \nwomen and adolescents within the reproductive age range of 15-49 years (1-3). However, among 36 \nwomen experiencing infertility, the prevalence of endometriosis can rise significantly, affecting 37 \napproximately 50% of this population. The disease can manifest as early as the first menstrual period 38 \nand persist until menopause, causing a wide range of symptoms that significantly impact quality of 39 \nlife. Notably, between 50% and 80% of women who suffer from chronic pelvic pain are diagnosed 40 \nwith endometriosis, highlighting its strong association with this symptom (1-3).  41 \nDespite its prevalence and impact, endometriosis remains underdiagnosed and poorly understood. 42 \nEarly detection and timely intervention are crucial for improving patient outcomes, yet diagnosis is 43 \nfrequently delayed by an average of 6 to 11 years (4, 5). This delay is attributed to several factors, 44 \nincluding the unknown etiology of the disease, the nonspecific and varied nature of its symptoms, 45 \nand the limitations of current diagnostic methods. Symptoms such as dysmenorrhea, chronic pelvic 46 \npain, dyspareunia, and gastrointestinal disturbances can often mimic other gynecologic or 47 \ngastrointestinal conditions, leading to misdiagnosis or delayed recognition. This variability, coupled 48 \nwith the lack of reliable diagnostic tools, poses significant challenges for both clinicians and 49 \nresearchers (6, 7).  50 \nThe current “Gold Standard” diagnostic endometriosis test involves an invasive laparoscopic surgical 51 \nprocedure, which carries risks of infection, bleeding, and damage to surrounding tissues (8-10). The 52 \nprocedure is only as reliable as the operator's surgical experience in locating and identifying 53 \nendometriosis lesions. As a result, its diagnostic accuracy varies greatly, with some studies reporting 54 \nsensitivity and specificity below 70% (10, 11). For instance, a study in Canada found that while 55 \nlaparoscopic visualization had a sensitivity of 90%, its specificity was only 40% (12). Another study 56 \nreported a sensitivity of 69% and a specificity of 83% for laparoscopy in diagnosing endometriosis 57 \n(13). These findings highlight that while laparoscopy is a valuable diagnostic tool, it is not infallible. 58 \nBesides relying heavily on the operator’s experience, the heterogeneous appearance of endometriotic 59 \nlesions precluded a standardized visual assessment, and the presence of deep infiltrating lesions may 60 \nbe missed during the procedure (10, 14). In addition, data suggest that some women undergoing the 61 \nprocedure do not have the condition and are therefore unnecessarily exposed to surgical risks. 62 \nFurther diagnostic advances may be expected from ever-evolving and indeed less invasive imaging 63 \ntechnologies, however as yet this has not achieved the gold standard status.   64 \nGiven these limitations, there is an increasing interest in developing non-invasive diagnostic methods 65 \nfor endometriosis (15-17). A growing body of evidence highlights the significant role of miRNA 66 \nexpression changes in the pathogenesis and progression of various diseases, including endometriosis. 67 \nNumerous research groups have focused on the regulatory role of miRNAs in key genes associated 68 \nwith endometriosis and have explored their potential as diagnostic biomarkers. For example, studies 69 \nhave identified miR-200 and miR-17-5p as promising biomarkers in the plasma of endometriosis 70 \npatients (18, 19), while miR-135a has shown diagnostic potential in the saliva of these patients (20). 71 \nHowever, conflicting results have also emerged regarding the diagnostic value of miRNAs (21). For 72 \nexample, one study found that serum miR-199a was upregulated in individuals with endometriosis 73 \n(22), while another study reported its downregulation in serum (23), highlighting the challenges in 74 \ninterpreting these findings and developing reliable diagnostic tests. Several factors may contribute to 75 \nthese discrepancies. For instance, the menstrual phase during which samples are collected may 76 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 4\ninfluence miRNA expression, as studies have shown that miRNA levels can vary between the 77 \nproliferative and secretory phases of the menstrual cycle (24, 25). Additionally, variations in study 78 \nprotocols, such as differing inclusion and exclusion criteria, may introduce further inconsistencies, 79 \nespecially considering the heterogeneity of endometriosis as a disease. Despite the numerous 80 \nchallenges associated with standardizing miRNA-based diagnostics, such as biological variability, 81 \ntechnical inconsistencies, and the complexity of endometriosis as a disease, considerable progress 82 \nhas been made in developing and implementing diagnostic tools in clinical settings. One notable 83 \nexample is the Endotest developed by Ziwig Inc., a test that utilizes a next-generation sequencing 84 \n(NGS) panel comprising over 100 microRNAs to evaluate disease status (26, 27). This test has been 85 \nprimarily introduced in European healthcare systems and represents a significant advancement in 86 \nnon-invasive diagnostics for endometriosis. Although the Ziwig Endotest holds substantial potential 87 \nas a laboratory-developed test (LDT), its broader implementation as a routine in vitro diagnostic 88 \n(IVD) tool remains limited. This is largely due to the high cost associated with NGS technologies 89 \nand the requirement for advanced bioinformatics infrastructure and expertise (28, 29), which may not 90 \nbe readily available in routine clinical settings. As such, while promising, NGS-based test is 91 \ncurrently better suited for use in a centralized reference laboratory rather than widespread clinical 92 \ndeployment in the form of IVD (30). 93 \nOne major obstacle in translating NGS-derived miRNA biomarkers into clinically actionable 94 \nPCR-based assays (qPCR or ddPCR) is the challenge of selecting appropriate endogenous controls 95 \nfor data normalization (31, 32). Unlike NGS, which employs global normalization strategies such as 96 \nreads per kilobase million (RPKM), and transcripts per million (TPM), PCR-based approaches 97 \nrequire specific endogenous reference genes for normalization. However, many studies rely on 98 \n\"universal\" endogenous controls, such as GAPDH, RNU48, or miR-16, without assessing their 99 \nstability in the specific disease contexts (31, 32). This oversight can introduce systematic bias, 100 \nleading to unreliable and non-reproducible biomarker quantification, thereby limiting their clinical 101 \nutility.   102 \nTo address these challenges, we conducted a proof-of-concept study integrating unbiased miRNA 103 \nbiomarker discovery via miRNA sequencing (miRNA-seq) with subsequent experimental validation 104 \nusing qPCR. Our primary objective is to present the development and feasibility of a miRNA-based 105 \npanel of biomarkers that would enable a non-invasive, simple and accurate blood test in diagnosing 106 \nendometriosis reliably. To control for female-specific physiological processes influenced by 107 \nhormonal fluctuations, such as those occurring during the menstrual cycle, we limited serum 108 \ncollection to women in the secretory phase. Another key innovation of our study is that we not only 109 \nidentified potential biomarkers but also established endometriosis-specific reference endogenous 110 \ncontrols for normalization in clinical diagnostic settings. These newly identified endogenous controls 111 \nwere systematically validated to ensure their stability and suitability for qPCR- and ddPCR-based 112 \nassays. Experimental validation of both the candidate biomarkers and the newly selected endogenous 113 \ncontrols demonstrated strong reproducibility and accuracy, supporting their potential clinical 114 \napplication. This approach represents a critical step toward the development of a standardized, 115 \nreliable, and clinically translatable miRNA-based diagnostic assay for endometriosis.116 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 5\nResults 117 \nDifferentially expressed miRNAs in endometriosis patients 118 \nIn the discovery cohort of samples collected during the secretory phases, we identified a total of 85 119 \ndifferentially expressed miRNAs between the patient group and the control group (Figure 1A; 120 \nSupplementary Table S1). Among these, 55 miRNAs were significantly upregulated in the disease 121 \ngroup, while 30 were downregulated, indicating a distinct molecular signature associated with 122 \nendometriosis. Interestingly, some subjects exhibited higher expression levels of specific miRNAs 123 \ncompared to others within the same group. This variability is likely attributable to the heterogeneous 124 \nnature of endometriosis, which can manifest differently across individuals in terms of severity, 125 \nsymptom presentation, and molecular profiles. Notably, several miRNAs previously implicated in the 126 \npathogenesis and progression of endometriosis were found to be highly dysregulated in the disease 127 \ngroup compared to the control group (Figure 1A-B). These included miR-9-5p and miR-21-5p, both 128 \nof which have been reported to play critical roles in regulating inflammatory signaling pathways, a 129 \nkey mechanism underlying the development and progression of endometriosis. 130 \nTo further investigate the origin and functional relevance of these differentially expressed miRNAs, 131 \nwe performed a hypergeometric over-representation analysis using miRNet (33, 34). This analysis 132 \nrevealed that the majority of these miRNAs predominantly originated from the bone marrow and 133 \ncervix (Figure 1C). The bone marrow association reflects the serum origin of the miRNA data. The 134 \ncervix, on the other hand, is relevant as it points to the anatomical regions where endometriosis 135 \nlesions can be found in proximity, providing confidence that the observed molecular signals are 136 \nindeed linked to the disease. This suggests that the serum circulating miRNAs may serve as 137 \nbiomarkers for systemic changes associated with endometriosis 138 \nAdditionally, we conducted a hypergeometric test using the gene targets of the differentially 139 \nexpressed miRNAs against the KEGG database to identify enriched molecular pathways (34). This 140 \nanalysis highlighted that the top non-tumor-specific molecular processes included various signaling 141 \npathways and focal adhesion, which are strongly associated with endometriosis (Figure 1C). For 142 \ninstance, focal adhesion is a critical process in the pathogenicity of endometriosis, as it involves 143 \ncell-matrix interactions that contribute to the attachment, survival, and invasion of endometrial cells 144 \noutside the uterus (35). Similarly, dysregulation in signaling pathways, such as the MAPK signaling 145 \npathway has been observed in endometrial cells, leading to enhanced cell proliferation, 146 \nendometriosis lesion establishment, and disease persistence (36). Furthermore, MAPK signaling is 147 \nimplicated in pain sensitization, suggesting a role in the chronic pelvic pain experienced by many 148 \npatients with endometriosis (37).  149 \n 150 \nAssessment of differentially expressed miRNAs for prediction of endometriosis using machine 151 \nlearning 152 \nTo evaluate whether the identified differentially expressed miRNAs possess predictive value in 153 \ndistinguishing individuals with endometriosis from control subjects, we implemented and assessed 154 \nthree distinct random forest models, each utilizing a different dataset.   155 \nFor the first model (Model #1), we constructed a random forest classifier using all 85 differentially 156 \nexpressed miRNAs. To rigorously assess its performance, we employed a 30-fold repeated 157 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 6\nsubsampling cross-validation approach. Model #1 demonstrated strong predictive capability, 158 \nachieving an overall sensitivity of 0.91, specificity of 0.88, and an area under the curve (AUC) of 159 \n0.95 (Figure 2A). To further refine the model and enhance interpretability, we explored feature 160 \nselection based on importance scores derived from the initial model. Specifically, we developed two 161 \nadditional models with reduced feature sets: Model #2, which utilized the top 40 most informative 162 \nmiRNAs, and Model #3, which incorporated only the top 20 (Supplementary Table S2). These 163 \nrefined models were constructed using the same random forest framework and evaluated with the 164 \nidentical repeated subsampling cross-validation procedure. By focusing on the most predictive 165 \nmiRNAs, we aimed to improve classification accuracy while eliminating noise introduced by less 166 \ninformative features.   167 \nThe results indicated that both reduced-feature models exhibited slightly improved performance 168 \ncompared to Model #1. Model #2, incorporating the top 40 miRNAs, achieved a sensitivity of 0.93, a 169 \nspecificity of 0.89, and an AUC of 0.97 (Figure 2B). Model #3, which was restricted to the top 20 170 \nmiRNAs, further enhanced predictive accuracy, yielding a sensitivity of 0.95, a specificity of 0.90, 171 \nand an AUC of 0.98 (Figure 2C). These findings suggest that a substantial proportion of the 172 \ndifferentially expressed miRNAs may not meaningfully contribute to disease classification and 173 \ninstead introduce noise into the predictive model. The improved performance of the reduced-feature 174 \nmodels underscores the value of feature selection in enhancing both the robustness and 175 \ninterpretability of machine learning-based biomarker discovery in endometriosis.   176 \n 177 \nSelection of miRNA biomarkers and endogenous controls for non-NGS clinical diagnostics 178 \nThe translation of NGS discovery findings into actionable clinical diagnostics is essential for 179 \nimproving patient care. While NGS provides comprehensive molecular insights, clinical diagnostic 180 \napplications often rely on qPCR due to their faster turnaround time, lower costs, and suitability for 181 \nIVD implementation without requiring the complex bioinformatics infrastructure necessary for NGS 182 \nanalysis (28-30). Given these advantages, we explored the feasibility of translating our NGS-based 183 \nfindings into a qPCR-based diagnostic assay. 184 \nOne of the primary considerations in this transition is the limit of detection of qPCR. The 85 185 \ndifferentially expressed miRNAs identified in our NGS analysis exhibit a wide range of expression 186 \nlevels across samples, spanning from as few as 10 normalized read counts (e.g., miR-4710) to over 187 \n1,000,000 (e.g., miR-21-5p) (Supplementary Table S1). Based on initial qPCR assessments, we 188 \ndetermined that reliable detection in qPCR is achieved for miRNAs with an average NGS normalized 189 \nread count exceeding 500. Applying this threshold, 38 of the differentially expressed miRNAs fell 190 \nbelow the cutoff and were deemed unreliable for qPCR-based detection (Figure 3A). This left 47 191 \ndifferentially expressed miRNAs that met the detection threshold for further investigation 192 \n(Supplementary Table S3).   193 \nTo assess the predictive potential of these 47 miRNAs in a qPCR setting, we constructed two random 194 \nforest models. The first model incorporated all 47 differentially expressed miRNAs, while the second 195 \nutilized a reduced feature set consisting of the 20 most informative miRNAs selected from this pool. 196 \nNotably, many of the miRNAs identified as top contributors in the previous analysis using all 85 197 \nmarkers—such as miR-21-5p, miR-17-5p, and miR-15b-5p—remained among the top-ranked features, 198 \nsuggesting that key predictive biomarkers identified via NGS may indeed be translatable to a clinical 199 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 7\nqPCR assay.   200 \nPerformance evaluation of these models revealed that the first model (using all 47 miRNAs) 201 \nachieved an AUC of 0.78 (Figure 3B), whereas the second model (using the top 20 miRNAs) 202 \nexhibited improved performance with an AUC of 0.84 (Figure 3C). As observed previously, reducing 203 \nthe feature set led to a slight increase in performance, likely due to the removal of uninformative or 204 \nnoisy variables. However, it is important to note that these qPCR-based models demonstrated lower 205 \noverall performance compared to the models trained on all 85 differentially expressed miRNAs in 206 \nthe NGS dataset. This suggests that some of the most informative miRNAs are expressed at low 207 \nlevels in the samples, such as miR-9-5p, which was ranked as the most predictive but had an average 208 \nnormalized read count of only ~350. The inability of qPCR to reliably detect such lowly expressed 209 \nmiRNAs presents a challenge for clinical translation. Addressing this limitation may require 210 \noptimizing primer designs, adopting ddPCR for enhanced sensitivity, or incorporating target capture 211 \nmethods to improve miRNA detection.   212 \nA second critical factor in transitioning to a qPCR-based assay is the selection of a suitable 213 \nendogenous control or reference marker for data normalization. Inappropriate reference markers can 214 \nintroduce variability, leading to unreliable expression quantification and potentially skewing 215 \ndiagnostic outcomes. To address this, we implemented a bioinformatic pipeline in identifying 216 \ndisease-specific endogenous controls (Ref. Materials and Methods). Briefly, we selected miRNAs 217 \nthat display minimal inter-group variability in expression to be used as endogenous references 218 \n(Supplementary Table S4). One such candidate, miR-92a-3p (Figure 3D), demonstrated consistent 219 \nexpression levels between individuals with endometriosis and control subjects. Notably, miR-92a-3p 220 \nhas been validated in prior studies as a reliable endogenous control in blood-based miRNA analyses 221 \n(38).   222 \nTo further refine our models, we performed in silico normalization of the 47 differentially expressed 223 \nmiRNAs against miR-92a-3p, simulating the clinical qPCR diagnostic workflow. Following 224 \nnormalization, we re-evaluated the performance of our random forest models. This approach led to 225 \nnoticeable improvements in predictive accuracy, with the model using all 47 miRNAs achieving an 226 \nAUC of 0.80, and the reduced model using the top 20 miRNAs attaining an AUC of 0.88 (Figures 227 \n3E-F). These findings underscore the importance of proper normalization strategies in enhancing the 228 \nrobustness of qPCR-based diagnostics and further support the potential clinical applicability of our 229 \nNGS-derived biomarker panel. 230 \n 231 \nExperimental assessment for diagnostic qPCR assays 232 \nTo evaluate the potential of translating our NGS discovery findings into a clinically viable qPCR 233 \ndiagnostic assay, we selected five candidate miRNAs — miR-21-5p, miR-15b-5p, miR-17-5p, 234 \nmiR-19b-3p, and miR-23a-3p — for experimental validation using qPCR. These miRNAs were 235 \nchosen based on their differential expression patterns observed in the NGS dataset, as well as their 236 \nbiological relevance to endometriosis. In addition, miR-92a-3p was included as an endogenous 237 \ncontrol for normalization, given its demonstrated stability across endometriosis and control samples 238 \nin both our dataset and prior studies.   239 \nFor this validation study, we conducted qPCR assays on 69 serum samples, comprising 45 patients 240 \nwith endometriosis and 24 control subjects whose disease status was confirmed via laparoscopic 241 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 8\nsurgery. The goal was to determine whether the expression patterns observed in the NGS discovery 242 \ndataset could be reliably recapitulated using qPCR, a crucial step in transitioning towards a clinically 243 \ndeployable assay. Analysis of the qPCR results revealed that two of the five selected biomarkers 244 \n(miR-21-5p and miR-15b-5p) displayed statistically significant differences in normalized expression 245 \nbetween the endometriosis and control groups. These findings are consistent with our NGS discovery 246 \ndataset, reinforcing their potential utility as diagnostic biomarkers. Additionally, miR-17-5p exhibited 247 \na discernible trend of differential expression between disease and control samples, as visualized in 248 \nthe boxplot analysis. However, this difference did not reach statistical significance, suggesting that 249 \nwhile this marker may hold some biological relevance, additional optimization—such as larger 250 \nsample sizes or refined qPCR conditions—may be necessary to establish its diagnostic value. In 251 \ncontrast, the remaining two miRNAs, miR-19b-3p and miR-23a-3p, did not show clear delineation 252 \nbetween patients and controls, indicating that their differential expression in the NGS dataset may 253 \nnot translate robustly into a qPCR-based assay. This discrepancy underscores the complexities of 254 \nbiomarker translation and highlights the need for rigorous validation and optimization at the 255 \nexperimental level. Factors such as primer design, amplification efficiency, RNA extraction 256 \nvariability, and technical noise could all contribute to differences between NGS and qPCR results, 257 \nemphasizing the importance of careful assay development. Notwithstanding, the data suggests that 258 \nqPCR-based assays could provide meaningful diagnostic utility with proper refinement. 259 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 9\nDiscussion 260 \nSmall RNAs such as miRNAs have been shown to play pivotal roles in numerous physiological and 261 \npathological processes, influencing gene regulation in both normal and disease states. Altered 262 \nmiRNA expression profiles have been extensively documented in the plasma and serum of patients 263 \nwith various conditions, including diffuse large B-cell lymphoma, ovarian cancer, and type 2 264 \ndiabetes (39, 40). These findings underscore the potential of circulating miRNAs as non-invasive 265 \nbiomarkers for disease detection and monitoring. In this study, we demonstrate that miRNAs 266 \ncirculating in serum can potentially serve as reliable biomarkers for the diagnosis of endometriosis. 267 \nThe ability to analyze serum miRNA levels in a standardized manner presents a promising approach 268 \nin disease detection. The fluctuations in specific circulating miRNAs offer a quantifiable and 269 \nreproducible means of identifying endometriosis, potentially improving early diagnosis and clinical 270 \nmanagement.   271 \nUsing serum miRNAs as diagnostic biomarkers offers several key advantages over conventional 272 \ndiagnostic methods in endometriosis. Currently, the gold standard for endometriosis diagnosis relies 273 \non laparoscopy with direct visualization, an invasive surgical procedure. A serum-based miRNA 274 \nbiomarker assay could provide a non-invasive alternative, enabling comprehensive disease 275 \nassessment without the need for surgery. This is particularly valuable for early detection and for 276 \npatients who may not have immediate access to specialized surgical evaluation. Secondly, compared 277 \nto invasive diagnostic procedures, a serum-based miRNA test is significantly more cost-effective. 278 \nThe process involves routine blood collection and standard laboratory processing, making it more 279 \naccessible for widespread clinical implementation. Additionally, standardizing miRNA detection 280 \nprotocols could facilitate large-scale screening efforts, improving early diagnosis and patient 281 \noutcomes.   282 \nIn this study, we investigated serum miRNA expression profiles in individuals with endometriosis 283 \nand identified a distinct set of circulating miRNAs that may serve as potential biomarkers for disease 284 \ndetection. To minimize variability associated with hormonal changes during the menstrual cycle, 285 \nparticularly those impacting female-specific physiological processes, serum samples were collected 286 \nexclusively during the secretory phase. Our results add to the growing evidence supporting the use of 287 \nserum miRNA signatures as non-invasive diagnostic tools for endometriosis (15-26). Nonetheless, 288 \ndespite their promise, several key challenges remain before miRNA-based assays can be translated 289 \ninto clinically reliable diagnostic applications. 290 \nOne of the primary obstacles lies in the choice of detection platform. While NGS provides a 291 \ncomprehensive assessment of the miRNA landscape, its high cost per sample and dependence on 292 \ncomplex bioinformatics infrastructure render it impractical for routine clinical diagnostics and IVD 293 \napplications. Consequently, there is a need to transition toward more practical methodologies, such 294 \nas qPCR or ddPCR, which offer lower costs, faster turnaround times, and compatibility with IVD 295 \nrequirements.  296 \nOur findings suggest that converting NGS-based miRNA discoveries into clinically applicable assays 297 \nis achievable, though it necessitates experimental refinement. Despite using a limited qPCR panel 298 \nwith only five miRNA biomarkers, we demonstrated valuable diagnostic potential, albeit with a need 299 \nfor further optimization. This highlights both the promise and the technical challenges of 300 \nimplementing miRNA-based diagnostics in clinical practice. 301 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 1 0\nTo improve the reliability and diagnostic accuracy of a qPCR-based test, several key strategies 302 \nshould be considered. (a) Primer Optimization: Refining primer designs to enhance amplification 303 \nefficiency and specificity, particularly for low-abundance miRNAs. (b) Adopting ddPCR for 304 \nEnhanced Sensitivity: ddPCR offers improved precision and sensitivity, making it a suitable 305 \nalternative for detecting low-expressed miRNAs that may be missed by qPCR. (c) Expanding 306 \nSample Size: Increasing the number of clinical samples analyzed will improve statistical power and 307 \nensure robustness across diverse patient populations. (d) Developing a Multiplex Assay: Creating a 308 \nmultiplexed qPCR panel would allow for the simultaneous detection of multiple miRNA biomarkers, 309 \nstreamlining workflow and improving diagnostic efficiency.   310 \nWhile initial qPCR validation of selected miRNAs shows promise, further refinement is necessary 311 \nand underway to enhance assay reproducibility and clinical performance. Future studies will focus on 312 \nvalidating these biomarkers in larger, independent cohorts, optimizing detection methods, and 313 \nstandardizing protocols to ensure reproducibility across clinical testing laboratories. These efforts 314 \nwill be critical in bridging the gap between high-throughput discovery research and real-world 315 \nclinical application, ultimately paving the way for a clinically deployable serum miRNA-based 316 \ndiagnostic test for endometriosis.   317 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 1 1\nMaterials and Methods 318 \nSpecimen collection 319 \nPeripheral blood samples were collected prospectively from women aged 18-49 years who presented 320 \nwith mild to severe symptoms, including pelvic pain and/or menstrual bleeding. The participants were 321 \nenrolled under an approved institutional review board protocol (IRB-20240110-R) at the Women’s 322 \nHospital of Zhejiang University School of Medicine, with informed consent obtained from all 323 \nindividuals. The participants included in this study were collected from March 2024 till December 324 \n2024. All participants were clinically suspected of a gynecologic abnormal condition and were 325 \nscheduled to undergo laparoscopy with histopathological evaluation. To account for potential 326 \nvariations in miRNA expression due to different menstrual cycle phases, blood samples were collected 327 \nexclusively from women in the secretory phase of their menstrual cycle. The menstrual phase was 328 \ninitially determined by physicians or surgeons based on self-reported cycle days and clinical 329 \nassessment. To ensure accuracy, this classification was further validated through serum progesterone 330 \nmeasurements using the protein assay from Kangrun Biotech Co. Ptd (Guangdong, China), with levels 331 \nexceeding 1.08 ng/mL serving as a biochemical confirmation of the secretory phase according to the 332 \nmanufacturer’s protocol. This approach minimized the influence of hormonal fluctuations on 333 \nbiomarker expression, thereby enhancing the reliability of our findings. A total of 40 symptomatic 334 \nwomen were included in the NGS discovery cohort, with 10 mL of blood drawn into standard red-top 335 \nblood collection tubes prior to the laparoscopic surgery. Among them, 20 women were confirmed to 336 \nhave endometriosis based on both laparoscopic findings and histopathology (disease group), while the 337 \nremaining 20 had no evidence of endometriosis and served as the control group. Serum was isolated 338 \nusing a two-step centrifugation protocol. First, samples underwent a low-speed centrifugation at 3000 339 \nrpm for 10 minutes at 4°C to remove cellular components. This was then followed by a second 340 \nhigh-speed centrifugation at 16,000 g for 10 minutes at 4°C to ensure complete removal of debris and 341 \nplatelets. The isolated serum was then aliquoted and stored at -80°C for subsequent RNA extraction 342 \nand downstream processing. 343 \n 344 \nRNA isolation and miRNA sequencing (miRNAseq)  345 \nTotal RNA was isolated from 300 μ L of serum using Norgen RNA extraction kit following the 346 \nmanufacturer's instructions. Total RNAs were ligated to 3’ adapters by denaturation at 70/i3  for 2 347 \nminutes, and then incubated at 16/i3  for over 8 hours using NEB T4 RNA Ligase 2. Following 5’ 348 \nadapters were incubated with previous product using NEB T4 RNA Ligase 1 at 37/i3  for 60 minutes. 349 \nLigated RNA was reverse transcribed in a thermocycler using SuperScript II Reverse Transcriptase 350 \nfrom ThermoFisher Inc. under the following conditions: an initial incubation at 50/i3  for 60 minutes, 351 \nfollowed by a heat inactivation step at 80/i3  for 10 minutes. Following cDNA synthesis, library 352 \npreparation was performed using the NEB Phusion High-Fidelity DNA Polymerase, adhering strictly 353 \nto the manufacturer's guidelines. The final libraries were then subjected to high-throughput 354 \nsequencing to profile the miRNA expression by LC Biosciences.  355 \n 356 \nNGS miRNA differential expression profiling endogenous control selection  357 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 1 2\nThe raw FASTQ data obtained from miRNA sequencing underwent preprocessing and analysis using 358 \nthe miRge3 (41) software pipeline. Initially, the sequences were quality-trimmed and adapter 359 \nsequences were removed using a Cutadapt (42) wrapper integrated within miRge3. The trimmed 360 \nreads were then aligned to the miRBase (43) reference database using Bowtie (44) optimized for 361 \nshort reads. Following alignment, miRge3 generated a comprehensive count table summarizing the 362 \nabundance of each miRNA across the samples. This count table served as the input for downstream 363 \ndifferential expression analysis using the DESeq2 (45) package in R. DESeq2 was employed to 364 \nidentify miRNAs that exhibited statistically significant differences in expression between the patient 365 \nand control groups. After identifying the differentially expressed miRNAs, hypergeometric 366 \nover-representation analyses were conducted using the miRNet platform (33). In this analysis, two 367 \ndistinct queries were performed: first, the differentially expressed miRNAs were compared against 368 \nthe miRNA-tissue origin database in miRNet to infer potential tissue-specific origins or associations 369 \nof these miRNAs; second, the predicted gene targets of these miRNAs were mapped to the KEGG 370 \n(Kyoto Encyclopedia of Genes and Genomes) database (34) to identify statistically enriched 371 \nbiological pathways. This dual-level approach enabled the contextualization of the miRNA 372 \nexpression patterns in terms of both tissue relevance and functional pathway involvement. 373 \n 374 \nEndogenous control selection for in silico normalization and qPCR experimental validation 375 \nTo bridge the findings from the next-generation sequencing (NGS) discovery cohort into clinically 376 \napplicable diagnostic tools, we further aimed to identify condition-specific endogenous control 377 \nmiRNAs. These controls are essential for normalizing quantitative PCR (qPCR) or droplet digital 378 \nPCR (ddPCR) assays, ensuring accurate and reproducible quantification of target miRNA expression. 379 \nThe selection of endogenous controls was guided by stringent criteria. Specifically, we evaluated the 380 \nexpression stability of candidate miRNAs by assessing their dispersion estimates and imposing 381 \nconstraints on log2 fold-change (|log2FC| < 0.02) between the patient and control groups. This 382 \nensured that the selected controls exhibited minimal variability across conditions. Additionally, 383 \ncandidates were filtered based on an adjusted p-value threshold (≥  0.8), ensuring that their expression 384 \nwas not influenced by the experimental conditions or disease state.  385 \n 386 \nDisease prediction model construction using a random forest classifier 387 \nTo assess the predictive capability of the differentially expressed miRNAs and determine the extent to 388 \nwhich they could accurately classify patients with endometriosis, we constructed a random forest 389 \nclassifier using all the differentially expressed miRNAs identified in the discovery cohort. To ensure a 390 \nrobust evaluation of the model's performance, we performed 30 iterations of repeated random 391 \nsubsampling cross-validation, where in each iteration, the data was split into an 80:20 ratio for training 392 \nand testing, respectively. This repeated holdout validation approach allowed us to account for 393 \nvariability in model performance due to random data partitioning and provided a reliable estimate of 394 \nthe model's predictive accuracy. During model construction, any missing values were imputed using 395 \nthe median of the corresponding feature. 396 \n 397 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 1 3\nFollowing the initial model construction, we conducted feature selection to identify the most 398 \ninformative miRNAs for predicting endometriosis. This was achieved by evaluating the feature 399 \nimportance scores generated by the random forest algorithm. Based on these scores, we created two 400 \ndistinct feature sets: one comprising the top 40 most important miRNAs and another consisting of 401 \nthe top 20 most important miRNAs. These reduced feature sets were then used to generate and assess 402 \nnew models using the same repeated random subsampling cross-validation procedure described 403 \nabove. Reducing the feature set is crucial for the assessment of model performance and 404 \ninterpretability. A smaller, more informative subset of miRNAs helps prevent overfitting, ensuring 405 \nthe model generalizes well to new data. Additionally, focusing on the most important miRNAs 406 \nreduces noise, leading to more reliable predictions. A reduced feature set also facilitates translation 407 \ninto clinical diagnostics where the informative miRNAs can be assayed using qPCR and/or ddPCR 408 \ninstead of NGS. 409 \n 410 \nRNA extraction and qPCR experimental validation 411 \nTotal RNA was extracted from 200 μ L serum using the miRNeasy Serum/Plasma Advanced Kit from 412 \nQiagen, following the manufacturer's recommended protocol. Subsequently, targeted microRNAs 413 \n(miRNAs) were reverse transcribed into complementary DNA (cDNA) using the FastKing RT Kit II 414 \nfrom TianGen Inc. The reverse transcription reaction was carried out in a thermocycler under the 415 \nfollowing conditions: an initial incubation at 42C for 15 minutes to facilitate cDNA synthesis, 416 \nfollowed by a heat inactivation step at 95C for 1 minute to terminate the reactions. The synthesized 417 \ncDNA was subjected to quantitative PCR analysis of the five miRNA markers. The PCR reaction 418 \nmixtures were prepared by miRCURY LNA miRNA SYBR Green PCR Kit from Qiagen. qPCR 419 \namplification was performed in QuantStudio qPCR system following: an initial denaturation at 95/i3  420 \nfor 2 minutes, followed of 40 cycles with denaturation at 95 /i3  for 10 seconds and annealing at 56/i3  421 \nfor 60 seconds. 422 \n 423 \nStatistical analyses 424 \nDifferential expression analysis was performed using DESeq2, which models NGS count data with a 425 \nnegative binomial distribution to assess statistical differences between patients and controls. For 426 \nmachine learning-based predictions, sensitivity, specificity, and AUC were evaluated using Python’s 427 \nscikit-learn package. Pairwise expression comparisons between patients and controls were assessed 428 \nusing the Wilcoxon ranked-sum test. 429 \n 430 \nStudy approval 431 \nThe participants were enrolled under an approved institutional review board protocol 432 \n(IRB-20240110-R) at the Women’s Hospital of Zhejiang University School of Medicine, with 433 \ninformed consent obtained from all individuals. 434 \n 435 \nData availability 436 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 1 4\nSequencing data were deposited into Genome Sequence Archive under the accession number 437 \nPRJCA039165438 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 1 5\nAuthor Contributions 439 \nY . Y u and W .H. Wong led the research group, analyzed the data and wrote the manuscript. Y . Yu, 440 \nW.H. Wong, X. Zhang and F. Bischoff conceptualized the study. X. Zhang and L. Zhu provided the 441 \nsamples while S. Lu coordinated sample collection between the hospital and the laboratory. W.H. 442 \nWong and Y . Hu performed random forest analysis. Y . Y u, Y . Shen and X. Xu performed molecular 443 \nexperiments. 444 \n 445 \nAcknowledgments 446 \nThe authors wish to express their gratitude to Jonathan Zhao and Frank Zhang (both from Heranova 447 \nLifesciences) for their valuable input on the study design and contributions to the development of the 448 \nmanuscript.  449 \n 450 \nConflict-of-interest statement: 451 \nAll authors, except Zhang Xinmei and Zhu Libo, are employees of Heranova Lifesciences, a 452 \ncompany engaged in the commercial development of a non-invasive test for endometriosis. Farideh 453 \nBischoff holds stock options of Heranova Lifesciences. The authors declare no other conflicts of 454 \ninterest, financial or otherwise.  455 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 1 6\nReferences 456 \n1. Zondervan KT, et al. Endometriosis. NEJM. 2020;382:1244-1256 457 \n2. Bonavina G, Taylor HS. Endometriosis-associated infertility: From pathophysiology to 458 \ntailored treatment. Front Endocrinol. 2022;13:1020827 459 \n3. Parasar P , et al. Endometriosis: Epidemiology, diagnosis and clinical management. Curr 460 \nObstet Gynecol Rep. 2017;6(1):34-41 461 \n4. Fryer J, et al. Understanding diagnostic delay for endometriosis: A scoping review using the 462 \nsocial-ecological framework. Health Care for Women International. 2025;46:335-351 463 \n5. Kirk UB, et al. Understanding endometriosis underfunding and its detrimental impact on 464 \nawareness and research. npj Women’s Health. 2024;2:45 465 \n6. Davenport S, et al. Barriers to a timely diagnosis of endometriosis: A qualitative systematic 466 \nreview. Obstet Gynecol. 2023;142:571-583 467 \n7. Pascoal E, et al. Strengths and limitations of diagnostic tools for endometriosis and relevance 468 \nin diagnostic test accuracy research. Ultrasound Obstet Gynecol. 2022;60(3):309-327 469 \n8. Leyland N, et al. Endometriosis: Diagnosis and management. J Obstet Gynaecol Can. 470 \n2010;32:S1-S32 471 \n9. Dunselman GA, et al. ESHRE guideline: Management of women with endometriosis. Hum 472 \nReprod. 2014;29:400-412 473 \n10. Taylor HS, et al. An evidence-based approach to assessing surgical versus clinical diagnosis 474 \nof symptomatic endometriosis. Int J Gynecol Obstet. 2018;142:131-142 475 \n11. Albee RB Jr, et al. Laparoscopic excision of lesions suggestive of endometriosis or otherwise 476 \natypical in appearance: Relationship between visual findings and final histologic diagnosis. J 477 \nMinim Invasive Gynecol. 2008;15:32-37 478 \n12. Gratton S, et al. Diagnosis of endometriosis at laparoscopy: A validation study comparing 479 \nsurgeon visualization with histologic findings. J Obstet Gynaecol Can. 2022;44:135-141 480 \n13. Sutton CJ, et al. Prospective, randomized, double-blind, controlled trial of laser laparoscopy 481 \nin the treatment of pelvic pain associated with minimal, mild, and moderate endometriosis. 482 \nFertil Steril. 1994;62:696-700 483 \n14. Afors K, et al. Employing laparoscopic surgery for endometriosis. Women’s Health. 484 \n2014;10:431-443 485 \n15. Vanhie A, et al. Plasma miRNAs as biomarkers for endometriosis. Hum Reprod. 486 \n2019;34:1650-1660 487 \n16. Avery JC, et al. Noninvasive diagnostic imaging for endometriosis part 1: A systematic 488 \nreview of recent developments in ultrasound, combination imaging, and artificial intelligence. 489 \nFertil Steril. 2024;121:164-188 490 \n17. Oskouei BS, et al. Non-invasive blood tests for earlier diagnosis and treatment of 491 \nendometriosis. J Reprod Immun. 2025;169:104521 492 \n18. Rekker K, et al. Circulating miR-200-family micro-RNAs have altered plasma levels in 493 \npatients with endometriosis and vary with blood collection time. Fertil Steril. 494 \n2015;104:938-946 495 \n19. Papari E, et al. Identification of candidate microRNA markers of endometriosis with the use 496 \nof next-generation sequencing and quantitative real-time polymerase chain reaction. Fertil 497 \nSteril. 2020;113:1232-1241 498 \n20. Perricos A, et al. Hsa-mir-135a shows potential as a putative diagnostic biomarker in saliva 499 \nand plasma for endometriosis. Biomolecules. 2022;12:1144 500 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 1 7\n21. Vanhie A, et al. Circulating microRNAs as non-invasive biomarkers in endometriosis 501 \ndiagnosis – a systematic review. Biomedicines. 2024;12:888  502 \n22. Wang WT, et al. Circulating microRNAs identified in a genome-wide serum microRNA 503 \nexpression analysis as noninvasive biomarkers for endometriosis. J Clin Endocrinol Metab. 504 \n2013;98:281-289 505 \n23. Hsu CY , et al. miRNA-199a-5p regulates VEGFA in endometrial mesenchymal stem cells and 506 \ncontributes to the pathogenesis of endometriosis. J Pathol. 2014;232:330-343 507 \n24. Petracco R, et al. Evaluation of miR-135a/b expression in endometriosis lesions. Biomedical 508 \nReports. 2019;11:181-187 509 \n25. Lamon S, et al. The effect of the menstrual cycle on the circulating microRNA pool in human 510 \nplasma: A pilot study. Hum Reprod. 2022;38:46-56 511 \n26. Bendifallah S, et al. Machine learning algorithms as new screening approach for patients with 512 \nendometriosis. Sci Rep. 2022;12:639 513 \n27. Ferrier C, et al. Saliva microRNA signature to diagnose endometriosis: A cost-effectiveness 514 \nevaluation of the Endotest. BJOG. 2023;130:396-406 515 \n28. Zalis M, et al. Next-generation sequencing impact on cancer care: Applications, challenges, 516 \nand future directions. Frontiers in Genetics. 2024;15:1420190 517 \n29. Desai K, et al. Real-world trends in costs of next-generation sequencing (NGS) testing in U.S. 518 \nsetting. Journal of Clinical Oncology. 2021;39:15_suppl 519 \n30. Karlovich CA, Williams PM. Clinical applications of next-generation sequencing in precision 520 \noncology. Cancer J. 2019:25:264-271 521 \n31. Chapman JR, Waldenstrom J. With reference to reference genes: A systematic review of 522 \nendogenous controls in gene expression studies. PLoS ONE. 2015;10(11):e0141853 523 \n32. Taylor SC, et al. The ultimate qPCR experiment: Producing publication quality, reproducible 524 \ndata the first time. Trends in Biotechnology. 2019;37(7):761-774 525 \n33. Chang L, et al. miRNet 2.0: Network-based visual analytics for miRNA functional analysis 526 \nand systems biology. Nucleic Acids Research. 2020;48:W244-W251 527 \n34. Kanehisa M, Goto S. KEGG: Kyoto encyclopedia of genes and genomes. Nucleic Acids 528 \nResearch. 2000;28(1):27-30 529 \n35. Nagai T, et al. Focal adhesion kinase-mediated sequences, including cell adhesion, 530 \ninflammatory response, and fibrosis, as a therapeutic target in endometriosis. Reproductive 531 \nSciences. 2020;27:1400-1410 532 \n36. Bora G, Yaba A. The role of mitogen-activated protein kinase signaling pathway in 533 \nendometriosis. J Obstet Gynaecol Res. 2021;47(5):1610-1623 534 \n37. Uimari O, et al. Genome-wide genetic analyses highlight mitogen-activated protein kinase 535 \n(MAPK) signaling in the pathogenesis of endometriosis. Human Reproduction. 536 \n2017;32(4):780-793 537 \n38. Solayman MHM, et al. Identification of suitable endogenous normalizers for qRT-PCR 538 \nanalysis of plasma microRNA expression in essential hypertension. Mol Biotechnol. 539 \n2016;58(3):179-187 540 \n39. Leva GD, Croce CM. miRNA profiling of cancer. Curr Opin Genet Dev. 2013;23(1):3-11 541 \n40. Vasu S, et al. MicroRNA signatures as future biomarkers for diagnosis of diabetes states. 542 \nCells. 2019;8:1533 543 \n41. Patil AH, Halushka MK. miRge3.0: A comprehensive microRNA and tRF sequencing 544 \nanalysis pipeline. NAR Genomics and Bioinformatics. 2021;3(3):lqab068 545 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 1 8\n42. Marcel M. Cutadapt removes adapter sequences from high-throughput sequencing reads. 546 \nEMBnet.journal. 2011;17(1):10-12 547 \n43. Kozomara A, et al. miRbase: From microRNA sequences to function. Nucleic Acid Research. 548 \n2018;47:D155-D162 549 \n44. Langmean B, et al. Ultrafast and memory-efficient alignment of short DNA sequences to the 550 \nhuman genome. Genome Biology. 2009;10:R25 551 \n45. Love MI, et al. Moderated estimation of fold change and dispersion for RNA-seq with 552 \nDESeq2. Genome Biology. 2014;15:550 553 \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n19\n554 \nFigure 1: Differential expression analysis of miRNAs in endometriosis patients. 555 \n(A) Heatmap illustrating the expression profiles of miRNAs that were differentially expressed 556 \nbetween endometriosis patients and healthy controls. A total of 85 miRNAs showed significant 557 \nexpression differences. Notably, several of these miRNAs, including miR-9-5p, miR-21-5p, 558 \nand let-7b-3p, have been previously associated with the pathogenesis of endometriosis and are 559 \nhighlighted on the plot. (B) Volcano plot depicting the distribution of differentially expressed miRNAs560 \nbased on fold change and statistical significance. (C) Pathway enrichment analysis of the differentially561 \nexpressed miRNAs. The top panel shows inferred tissue-of-origin patterns, suggesting potential source562 \ntissues of the dysregulated miRNAs. The bottom panel presents enriched non-cancer-related KEGG 563 \npathways derived from the predicted target genes of these miRNAs, highlighting relevant biological 564 \nprocesses potentially involved in endometriosis pathophysiology. 565 \n \nAs \nlly \nrce \nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 2 0\n 566 \nFigure 2: Diagnostic performance of NGS-identified differentially expressed miRNAs using 567 \nmachine learning. 568 \nThe diagnostic utility of differentially expressed miRNAs was assessed through machine-learning 569 \nanalysis. Performance metrics represent the mean of 30 iterations of repeated subsampling 570 \ncross-validation, with error bars indicating variability across iterations. (A) Using the full set of 571 \ndifferentially expressed miRNAs, the model achieved an AUC of 0.95, with a sensitivity of 0.91 and 572 \nspecificity of 0.88. A representative ROC curve from iteration #10 is shown. (B) Limiting the model to 573 \nthe top 40 most informative miRNAs improved performance, yielding an AUC of 0.97, sensitivity of 574 \n0.93, and specificity of 0.89. The ROC curve displayed corresponds to iteration #12. (C) Further 575 \nrefinement using the top 20 most informative miRNAs resulted in the highest diagnostic performance, 576 \nwith an AUC of 0.98, sensitivity of 0.95, and specificity of 0.90. The representative ROC curve is also 577 \nfrom iteration #12. 578 \nA B C\nAUC Sensit ivit y Speci /i1cit y\n0.00\n0.25\n0.50\n0.75\n1.00\n0.00\n0.25\n0.50\n0.75\n1.00\nAUC Sensitivit y Speci\n/i1c ity A U C S e ns itivity S pe c i /i1cit y\n0.00\n0.25\n0.50\n0.75\n1.000.95 0.91 0.88\n0.97 0.93 0.89\n0.98 0.95 0.90\nCross Validation #10 (Representative)\n0.2\n0.4\n0.6\n1.0\n0.8\nT rue P os itive R ate\nF als e P os itive R ate\n0 . 20 . 4 0 . 60 . 81 . 0\nModel 1: All DE miRNAs Model 2: Top 40 DE miRNAs Model 2: Top 20 DE miRNAs\n0.2\n0.4\n0.6\n1.0\n0.8\nT rue P os itive R ate\n0.2\n0.4\n0.6\n1.0\n0.8\nT rue P os itive R ate\nF als e P os itive R ate\n0.2 0.4 0.6 0.8 1.0\nF als e P os itive R ate\n0 . 20 . 4 0 . 60 . 81 . 0\nAUC = 0.93 AUC = 0.94 AUC = 0.94\nCross Validation #12 (Represent ative) Cross Validat ion #12 (Representat ive)\nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 2 1\n 579 \nFigure 3: Diagnostic performance of differentially expressed miRNAs reliably quantifiable by 580 \nqPCR, assessed using machine learning. Performance metrics represent mean of 30 iterations of 581 \nrepeated subsampling cross-validation, with error bars indicating variability across iterations. (A) 582 \nViolin plot illustrating the expression distribution of differentially expressed miRNAs based on 583 \nnormalized NGS read counts. miRNAs shown in blue represent those with expression levels too low 584 \nfor reliable qPCR detection, while those in red indicate miRNAs with sufficient expression for reliable 585 \nquantification by qPCR. The dotted red line marks the expression threshold of 500 normalized reads, 586 \nused to distinguish between the two groups. (B) Predictive performance of a machine-learning model 587 \nusing all 47 miRNAs deemed reliably detectable by qPCR, resulting in an average AUC of 0.78, with 588 \na sensitivity of 0.66 and specificity of 0.69. (C) Model performance using the top 20 most informative 589 \nmiRNAs from the qPCR-detectable set, showing a modest improvement with an average AUC of 0.84, 590 \nsensitivity of 0.68, and specificity of 0.74. (D) Boxplot of miR-92a-3p, a potential endogenous control 591 \ncandidate in this disease setting, demonstrating consistent expression with minimal variability 592 \nbetween patient and control groups. The box’s lower and upper hinges correspond to the 25th and 75th 593 \npercentiles, respectively, with the median indicated by the line inside the box. The whiskers extend to 594 \nthe most extreme data points within 1.5 times the interquartile range below the 25th percentile and 595 \nabove the 75th percentile. (E) Predictive performance using all 47 reliably detectable miRNAs after in 596 \nsilico normalization against miR-92a-3p, yielding an average AUC of 0.80, with sensitivity of 0.70 597 \nand specificity of 0.68. (F) Model performance using the top 20 most informative miRNAs following 598 \nin silico normalization with miR-92a-3p, showing further improvement with an average AUC of 0.88, 599 \nsensitivity of 0.77, and specificity of 0.68. 600 \nA B C\nD E F\nDetectable \nvia qP CR\nNot det ectable \nvia qPCR\nlog2(average expres sion)\nDE miRNAs\n0\n5\n10\n15\n20\n0.00\n0.25\n0.50\n0.75\n1.00\nAUC Sensitivit y Speci /i1cit y\n0.00\n0.25\n0.50\n0.75\n1.00\nAUC S ensitivit y S peci /i1cit y\n0.00\n0.25\n0.50\n0.75\n1.00\nAUC Sensitivit y Speci\n/i1c ity A U C S e ns itivity S pe c i /i1cit y\n0.00\n0.25\n0.50\n0.75\n1.00\nAll qPCR Detectable miRNAs Top 20 qPCR Detectable miRNAs\nAll qPCR Det ect able miRNAs\nnormalized with miR-92a-3p\nTop 20 qPCR Det ect able miRNAs\nnormalized with miR-92a-3p\n0.78\n0.66 0.69\n0.84\n0.68\n0.74\n0.80\n0.70 0.68\n0.88\n0.77\n0.70\nC ontrols E ndometriosis\n500000\n750000\n1000000\n1250000Normalized Read Count\nmiR -92a-3p\nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint \n\n 2 2\n 601 \nFigure 4: Boxplot of qPCR performance for selected differentially expressed miRNAs. 602 \nAll miRNAs were normalized against miR-92a-3p using qPCR. Of the five miRNAs tested, two 603 \n(miR-21-5p and miR-15b-5p) showed a significant difference in expression between patients and 604 \ncontrols, replicating the direction of effect observed in the NGS discovery 605 \ndataset. miR-17-5p exhibited a concordant trend, with a p-value of 0.065, approaching statistical 606 \nsignificance. The remaining two miRNAs, miR-19b-3p and miR-23a-3p, did not show a statistically 607 \nsignificant difference between patients and controls. The boxplot’s lower and upper hinges represent 608 \nthe 25th and 75th percentiles, respectively, with the median indicated by the line inside the box. The 609 \nwhiskers extend to the most extreme data points within 1.5 times the interquartile range below the 25th 610 \npercentile and above the 75th percentile. 611 \nmiR-21-5p\nmiR-15b-5p\nmiR-17-5p\nmiR-19b-3p\nmiR-23a-3p\n2.7\n3.0\n3.3\n3.6\n-log2(Delta CT)\nmiR NAs normalized with miR -92a-3p\np = 0.0028 p = 0.0048 p = 0.0649 p = 0.3105 p = 0.4338\nCo n t r o l s\nn = 24\nE ndometriosis\nn = 45\n3.9\nAll rights reserved. No reuse allowed without permission. \n(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. \nThe copyright holder for this preprintthis version posted July 24, 2025. ; https://doi.org/10.1101/2025.07.23.25332110doi: medRxiv preprint","source_license":"CC0","license_restricted":false}