Interpretable machine-learning-derived diagnostic scoring panel for endometriosis identification: a study of serum amino acid profiling

Interpretable machine-learning-derived diagnostic scoring panel for endometriosis identification: a study of serum amino acid profiling | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Interpretable machine-learning-derived diagnostic scoring panel for endometriosis identification: a study of serum amino acid profiling Moyuan Li, Sujuan Xu, Yiwei Cao, Feiyang Li, Nuo Ye, Yiran Xu, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6696073/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Delayed diagnosis of endometriosis (EM) can significantly hinder treatment options and patient prognosis, adversely affecting fertility and increasing the risk of malignancy. Current diagnostic methods are invasive. In our study, ultra-high-performance liquid chromatography-tandem mass spectrometry-based serum amino acid profiling was used to compare amino acid abundance between 137 normal controls (NCs) and 167 patients with EM. Subsequently, five machine learning algorithms (random forest, neural network, extreme gradient boosting, support vector classification, and naïve Bayes) were used to establish the efficacy and stability of amino acid for diagnosing. Ultimately, seven amino acids, 3-methyl-L-histidine, kynurenine, leucine, N6-acetyl-L-lysine, phenylalanine, theanine, and tyrosine, were successfully identified as valuable variables using the SHapley Additive exPlanations (SHAP) method and were included in the diagnostic scoring panel. The areas under the receiver operating characteristic and precision-recall curves of the diagnostic scoring panel were greater than 0.8. Additionally, we characterized the EM group into several subclasses using K-Means clustering and analyzed differences between the subclasses. In conclusion, This study developed a non-invasive diagnostic scoring panel with excellent predictive ability. Consequently, it enhances the accuracy of EM diagnosis, mitigates the risk of malignancy, and facilitates the formulation of subsequent personalized treatment and prevention strategies. Endometriosis serum amino acids diagnostic scoring panel machine learning SHAP Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction Endometriosis (EM) is defined as the presence, growth, and infiltration of endometrial tissue (including glands and stroma) outside the uterine cavity, resulting in recurrent bleeding[ 1 ]. Ectopic endometrial glands undergo cyclic changes regulated by estrogen, leading to symptoms such as bleeding, pain, nodules or masses, and infertility, depending on the affected sites (including the ovaries, fallopian tubes, pelvic peritoneum, and pelvic ligaments)[ 2 , 3 ]. Previous studies indicate that 10–15% of women of reproductive age and 70% of those with chronic pelvic pain have EM. Furthermore, approximately 25–50% of infertile women have EM, and 30–50% of those with EM experience infertility[ 4 ]. Despite this prevalence, diagnosis is often delayed by an average of 6.7 years among individuals aged 18–45 years[ 5 ]. Such delays result in untimely treatment, adversely affecting fertility and increasing the risk of ovarian epithelial cancer, causing unnecessary suffering for women of reproductive age. Endometriosis diagnosis has traditionally relied on symptoms and signs, ultrasound examinations, and laparoscopic surgery. Ultrasound examination results are highly dependent on the operator’s skill, while laparoscopic surgery, the gold standard, is invasive, high cost, potentially damaging, and carries anesthesia risks[ 6 , 7 ]. Although serum cancer antigen 125 is commonly used in clinical practice as a non-invasive diagnostic method, studies show its relatively inadequate sensitivity and specificity[ 8 ]. Consequently, developing a non-invasive diagnostic method for EM has become a high priority. The mechanisms underlying EM are intricate, encompassing theories such as retrograde menstruation[ 9 ], implantation of endometrial stem cells[ 10 ], inflammatory factors[ 11 ], the immune microenvironment[ 12 ], and others[ 13 ], which significantly affect the occurrence and progression. Amino acids are small molecules essential in manifold physiological activities in the human body, including protein synthesis, physiological regulation, immune function, and energy supply. Moreover, disorders in amino acid metabolism are closely associated with the occurrence and progression of multiple inflammatory diseases, including EM[ 14 ]. Other studies have revealed that amino acids are potential metabolomic biomarkers for EM diagnosis[ 15 – 17 ]. However, the role of serum amino acid abundance combined with machine learning algorithms in the early diagnosis of EM remains limited. Machine learning is the scientific study of algorithms and models that aim to enhance efficiency and accuracy in clinical diagnosis and assist physicians in decision-making[ 18 ]. The novel combination of serum amino acid profiling and machine learning has the potential to identify correlates of endometriosis, facilitating early, non-invasive diagnosis and enhancing patient outcomes. Herein, we characterized the serum amino acid profile in EM, screened for potentially valuable variables using assorted machine learning algorithms, and established a diagnostic scoring panel for predicting incident EM using a logistic regression-derived nomogram model consisting of seven amino acids with high specificity, sensitivity, and selectivity. This study provides new insights and a scientific basis for early EM diagnosis, mechanistic research, and subsequent development of personalized treatment and prevention strategies. 2. Methods 2.1 Preparation of Standards and Establishment of Standard Curves In this study, 33 amino acids (3-methyl-L-histidine (3-MeHis), γ-aminobutyric acid (GABA), alanine (Ala), arginine (Arg), asparagine (Asn), aspartic acid (Asp), cystine (Cys), glutamic acid (Glu), glutamine (Gln), glycine (Gly), guanidinoacetic acid (Gua), histidine (His), homocitrulline (HCit), hydroxyproline (Hyp), isoleucine (Ile), kynurenine (Kyn), leucine (Leu), lysine (Lys), methionine (Met), N,N-dimethylglycine (Dmg), N6-acetyl-L-lysine (H-Lys(Ac)-OH), ornithine (Orn), phenylalanine (Phe), pipecolic acid (Pip), proline (Pro), pyroglutamic acid (PGA), S-adenosyl-L-homocysteine (SAH), serine (Ser), S-methyl-L-cysteine (SMC), theanine (The), threonine (Thr), tryptophan (Trp) and tyrosine (Tyr)) were purchased and prepared as a 5 mmol/L stock solution in 20% methanol aqueous solution, and stored at -80°C for future use. Precise aliquots of each standard stock mixture were stepwise diluted to prepare mixed standards for establishing the standard curve. 2.2 Multiple reaction monitoring (MRM) channel setup and Ultra-performance liquid chromatography-mass spectrometry (UPLC-MS/MS) conditions Standards for each substance were used to set the conditions for UPLC-MS/MS, including parent ion determination, parent ion fragmentation voltage optimization, daughter ion determination, daughter ion collision energy optimization, and MRM channel establishment. An Agilent 1260 liquid chromatography system coupled with a G6460 tertiary quadrupole mass spectrometer was used under the following chromatographic conditions: chromatographic column: Agilent Infinity Lab Poroshell 120 HIIC-Z (2.1×100 mm, 2.7 µm); flow rate of 0.5 mL/min; column temperature of 25°C, and; mobile phase A: 20 mmol/L ammonium formate acetonitrile solution (V:water:V: acetonitrile = 1:9); and mobile phase B: 20 mmol/L ammonium formate aqueous solution. Gradient elution was performed using the following linear gradient: A, 100% (0 min), 70% (11.5 min), 70% (12 min), and 100% (15 min). Mass spectrometry was performed in electrospray ionization positive-ion mode. 2.3 Patient recruitment and collection of clinical features Blood samples and clinical features were collected from 167 patients diagnosed with EM and 137 normal controls (NCs). These features include age, dysmenorrhea presence, blood test results conducted throughout the visit, including peripheral blood mononuclear cell (PBMC), cholesterol, triglycerides (TG), alanine aminotransferase (ALT), aspartate aminotransferase (AST), γ-glutamyl transpeptidase, high-density lipoprotein cholesterol, alpha-fetoprotein, carcinoembryonic antigen (CEA), cancer antigen 125 (CA 125), cancer antigen 199 (CA 199). A total of 167 patients with EM were diagnosed at Nanjing Medical University Affiliated Obstetrics and Gynecology Hospital between December 1, 2021, and December 1, 2023. All patients underwent laparoscopic examination with pathological confirmation of EM. The American Society for Reproductive Medicine (ASRM) stage, revised ASRM score, diameter, and whether the number of lesions were single or multiple were documented. Concurrently, 137 healthy female volunteers who underwent routine health checkups at our institution were recruited as the NC group. They had no chronic diseases, medication histories, or abnormalities on pelvic examination. The exclusion criteria were as follows: (1) pregnant or breastfeeding women; (2) aged 50 years; (3) history of medication treatment within the past 3 months; (4) other comorbidities such as hyperthyroidism, hypothyroidism, diabetes, or hypertension; and (5) refusal to participate or were uncooperative with the researchers. All participants provided written informed consent, and the research adhered to the Declaration of Helsinki. The ethics committee of the Nanjing Medical University Affiliated Gynecology Hospital approved all research protocols (No. 2022 KY-100). 2.4 Serum amino acids extraction Blood samples from the EM group were collected before surgery, while those from the NC group were collected after > 8 h of overnight fasting. The serum from blood samples was centrifuged again to obtain the supernatant, which was then transferred to an Eppendorf tube. Subsequently, 800 µL of pre-cooled methanol (containing 12.5 µmol/L of L-glutamate 13C5 solution) was added, and the mixture was vortexed vigorously for 5 min. Subsequently, the samples were incubated at -80°C for 8 h to ensure complete protein precipitation. Following incubation, the samples were centrifuged at 16,000 × g at 4°C for 15 min. The supernatant was carefully transferred to new tubes and lyophilized at 4°C until powdered. Before loading, the sample was re-dissolved in 100 µL of a solution containing 80% methanol and 20% water (methanol: water = 4:1) and further centrifuged for visualization. The supernatant was then transferred to a mass spectrometry vial. 2.5 Data processing Differences in clinical baseline characteristics were analyzed using the chi-square test for categorical data and the Student’s t-test (two-tailed) or one-way analysis of variance for continuous data. Agilent MassHunter Qualitative Analysis B.06.00 software was used to extract and analyze retention times, peak areas, and peak heights of 33 amino acids, normalizing data to an internal standard. Raw data underwent principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) using R, evidencing reproducibility within groups and differences between groups, thereby establishing a model correlating amino acid expression levels with group classification. Independent samples were compared using the Student’s t-test (two-tailed). In the heatmap and volcano plot analyses, filtered significantly up- and downregulated amino acids based on p 1.5 or < 0.67. The samples were subsequently categorized into training and testing sets at a 7:3 ratio. Thirty-three amino acids and 13 clinical indicators, including random forest (RF), neural network (NN), extreme gradient boosting, support vector classification (SVC), and naive Bayes, were used to establish multiple machine learning algorithms using Python software. The predictive performance of these models was evaluated using receiver operating characteristic (ROC) curves, precision-recall curves (PRC), and decision curve analysis (DCA). Given the clinical challenge of accepting predictions from models that lack interpretability, the SHapley Additive exPlanations (SHAP) method was utilized for the global interpretation of the machine learning algorithm results, highlighting the importance of features and selecting potentially valuable variables for constructing a diagnostic scoring panel. The selected biomarkers were analyzed using logistic regression analysis in R, and a nomogram was designed to represent disease risk prediction visually. The predictive performances of the five machine learning algorithms and the diagnostic scoring panel were evaluated using ROC curves, PRC, and DCA. The diagnostic score for each subject and total nomogram points were calculated from the nomogram model, and the ROC curve for total nomogram points determined the optimal threshold, classifying individuals with scores above the cutoff as high risk for EM. We characterized the EM group into several subclasses in terms of amino acid abundance using an unsupervised learning algorithm, K-Means Clustering. The uniform manifold approximation and projection (UMAP) algorithm subsequently demonstrated the efficacy of the clustering effect. Differences in the ASRM scores, diameters, and nomogram total points between subclasses were analyzed using the Wilcoxon test, while ASRM stage, single or multiple lesions, and percentage of dysmenorrhea were analyzed using Fisher’s exact test. 3. Results 3.1 Workflow Figure 1 illustrates the study workflow. First, metabolites were extracted from the samples using a liquid-liquid extraction process, followed by quantification of serum amino acids using the UPLC-MS/MS-based MRM method and variance analysis. Second, the abundance of serum amino acids was integrated with clinical features to establish predictive models using five machine-learning algorithms. The SHAP method selected potentially valuable variables, and a highly specific and sensitive diagnostic scoring panel was constructed before evaluating its effectiveness. Finally, all EMs were differentiated into three subclasses based on the abundance of amino acid expression, and correlations among these classes, clinical manifestations, and prognosis were identified. 3.2 Clinical characteristics of the participants This study included a cohort of 304 participants: 167 individuals diagnosed with EM and 137 NCs. The detailed demographic and clinical characteristics of all subjects are presented in Table S1 . Overall, differences in the number of peripheral PBMCs and TG, ALT, AST, and CEA levels were observed between the two groups. However, the levels of CA 125 and CA 199 in the EM group were significantly higher than those in the NC group. Additionally, patients with EM were more likely to experience dysmenorrhea than NC participants. No statistically significant differences in other demographic or clinical indicators were observed between the EM and NC groups ( P > 0.05). In the EM group, 84 single lesions were observed in the ovarian endometriosis (OvE) and 22 in the adenomyosis. Among multiple lesions, 48 were found in the OvE and 13 in the OvE combined with adenomyosis. 3.3 Establishment and validation of the amino acid’s quantification method We performed UPLC-MS/MS-based serum amino acid profiling. MS/MS spectrometric data, including parent and fragment ion information for each amino acid, were selected, and optimization of the decluttering potential and collision energy facilitated the establishment of the MRM method, with chromatograms of 33 amino acids shown in Fig. 2 A. The retention time for each metabolite was determined using a mixed standard solution of amino acids. Linear gradient elution effectively separated 33 amino acids within 10 min, with peak times primarily concentrated between 2 and 8 min. Excellent linearity was observed across various amino acid concentration ranges (Fig. 2 B). Table S2 showed the regression equation, r2 and linear range for 33 amino acids. 3.4 Identification of differentially abundant amino acids Using the established UPLC-MS/MS method, the retention times, peak areas, and peak heights of the amino acids were extracted and normalized, yielding corresponding levels for each sample. The PCA and OPLS-DA results revealed a separation trend between the EM and NC groups, with Q² = 0.735 (Fig. 3 A and 3 B). Statistical analysis revealed a significant difference in the abundance of 28 amino acids (p 1.5 or < 0.67 and a P < 0.05 were considered differentially abundant. The abundances of 3-MeHis, Kyn, Leu, H-Lys(Ac)-OH, Phe, The, Tyr, etc., were greater in EM samples than in the NC samples, except for Tyr, which was lower. Hierarchical cluster analysis revealed intragroup correlations and significant intergroup differences (Fig. 3 C). These amino acids are visible in the volcano plot (Fig. 3 D). The abundance of Kyn, Phe, Leu, and Tyr consistently differed between the two groups. Furthermore, analysis of the area under the ROC curve for each amino acid indicated that most had the potential to identify EM (Fig. 3 E). A trend of smaller P-values correlating with larger areas under curve values was observed, suggesting that differences in amino acid abundance enhanced predictive performance. 3.5 Prediction model development and valuable variables selection This study incorporated 33 amino acids and 13 clinical indicators as features to generate five machine-learning algorithms for assessing the risk of developing EM. The predictive performances of these five models in the training and test sets are shown in Figs. 4 A-C. The ROC curve analysis (Fig. 4 A) and PRC analysis (Fig. 4 B) indicated that the area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC) for all machine learning models in the training set was greater than 0.9, while in the testing set, they were both greater than 0.7. In the DCA, the net benefit curves of all machine learning models exhibited increased net benefit within a certain threshold probability range, coupled with a concomitant smooth transition (Fig. 4 C). Generally, the SVC and NN models showed greater net benefits, although the predictive performances of the other models are noteworthy and should not be overlooked. The SHAP method was employed to select valuable variables by calculating the contribution of each variable. As shown in Figure S2 , the average SHAP values were used to assess the contribution of the variables to each model, which are displayed in descending order. For example, it ranked first in its contribution to the overall function of the RF model in training and testing sets. As shown in Fig. 4 D, the intersection of the top 20 features across models revealed seven common features: 3-MeHis, Kyn, Leu, H-Lys(Ac)-OH, Phe, The, and Tyr. 3.6 Constructing a diagnostic scoring panel for predicting incident EM Through global interpretation using SHAP values, we identified the top seven common features as valuable variables for early EM diagnosis. A diagnostic scoring panel for predicting incident EM was constructed using a logistic regression nomogram (Fig. 5 A). This panel yielded AUROC values of 0.96 and 0.89 for the training and testing sets, respectively (Fig. 5 B), coupled with AUPRC values of 0.94 and 0.87, respectively (Fig. 5 C). The DCA results revealed satisfactory potential clinical benefits (Fig. 5 D). These findings emphasize the excellent predictive performance of the diagnostic scoring panel. Furthermore, 1,000 bootstrap resampling iterations with replacement were conducted, and the calibration curve demonstrated the robustness of the diagnostic scoring panel (Fig. 5 E). These findings emphasize the high predictive performance of the diagnostic scoring panel. Finally, each feature’s score was summed to obtain a total nomogram point, with a cutoff value of 159.86 determined by the Youden index, allowing for the classification of high-risk (nomogram total point > cutoff value) and low-risk (nomogram total point < cutoff value) groups (Fig. 5 F and 5 G). The heatmap revealed that the abundance of most amino acids, except tyrosine, was greater in the high-risk group, suggesting that tyrosine was a low-risk feature while others were high-risk. The total nomogram points in the EM group were significantly greater than those in the NC group ( Fig. S3A ), and this difference was statistically significant. A confusion matrix was obtained based on the classification of high- and low-risk groups. In the training set, the true-positive rate for predicting incident EM was 115/116 (99.14%), and the true-negative rate was 86/96 (89.58%). In the testing set, the true positive rate for predicting incident EM was 49/51 (96.08%), and the true negative rate was 31/41 (75.61%) ( Fig. S3B ). These results strongly indicate that this diagnostic scoring panel has significant value and clinical applicability for early diagnosis of EM. 3.7 Subclass analysis and progression prediction The UMAP clustering effect (Fig. 6 A) categorized the EM group into three subclasses based on amino acid abundance. The proportion of patients with stage IV ASRM (Fig. 6 B) and ASRM scores (Fig. 6 C) was significantly greater in class 3 than in the other two classes (p < 0.05), indicating more extensive lesions and severe symptoms. From the diagnostic scoring panel’s perspective, Class 3 patients had higher total nomogram points (Fig. 6 D, P < 0.05) and a greater risk of developing EM. Additionally, the diameter ( Fig. S4A ), presence of dysmenorrhea ( Fig. S4B ), and single or multiple proportions ( Fig. S4C ) did not differ among the three subclasses. Based on these results, we can predict disease progression using subclass analysis and provide ideas for managing surgery and personalized treatment. 4. Discussion This study used UPLC-MS/MS to determine the abundance of amino acids in the serum samples of patients with EM. In the variance analysis, 11 amino acids were significantly increased, whereas Tyr levels decreased in the EM group. Five machine-learning algorithms were subsequently developed using the SHAP method for potentially valuable variables. Finally, a diagnostic scoring panel comprising seven amino acids (3-MeHis, Kyn, Leu, H-Lys(Ac)-OH, Phe, The, and Tyr) was established using a logistic regression-derived nomogram model, demonstrating excellent potential for early EM diagnosis. This analysis identified specific amino acids that could serve as biomarkers for the early diagnosis of EM. Currently, non-invasive and potentially valuable variables for the early diagnosis of EM are still lacking. Patients with EM often experience symptoms such as bleeding, dysmenorrhea, abdominal masses, and infertility[ 2 ]. This condition severely affects women of childbearing age in China. However, diagnosis is frequently delayed, which affects prognosis and increases the risk of malignancy. However, laparoscopic and ultrasonographic examinations have limitations. Consequently, research efforts focus on developing non-invasive detection methods, such as liquid testing (blood, urine, and follicular fluid), to improve early diagnosis accuracy. CA125 is the most studied and used biomarker for EM, but its sensitivity and specificity are inadequate[ 8 ], showing greater efficacy in diagnosing advanced (stage III-IV) than early (stage I-II) endometriosis[ 19 ]. To date, over 100 potential EM biomarkers have been reported, but few are applicable to mild EM detection[ 20 ]. Therefore, applying machine-learning algorithms to identify reliable and consistent biomarkers is crucial. This approach established an early EM diagnostic model, helping to reduce infertility, cancer risk, and complications associated with late-stage diagnosis. Disruptions in amino acid metabolism are closely associated with EM progression, metabolic disorders, immune disorders, and cancer. Ectopic endometrial cells exhibit many similarities to cancer cells, including uncontrolled proliferation, cell invasion, neovascularization, and apoptosis resistance, as is widely recognized in the literature[ 21 , 22 ]. Therefore, it is hypothesized that the changes in amino acids during the progression of EM and cancer may be parallel. Traditionally, cancer metabolism has focused on central carbon metabolism, including glycolysis and the tricarboxylic acid cycle. However, recent studies have revealed that amino acids also play fundamental roles in energy regulation, biosynthetic support, and redox balance[ 23 ]. Similar to tumors, alterations in amino acid metabolism can serve as clinical indicators of EM progression and inform treatment strategies. Our study detected a significant increase in the levels of several amino acids, notably Leu, Phe, Ala, and Lys, in the serum of patients with EM compared to normal NCs. These changes may be attributed to physiological mechanisms related to tissue damage and biosynthetic repair associated with EM. In support of this notion, Dutta et al. suggested that the catabolic state induced by tissue damage in EM leads to an increased breakdown of endogenous proteins and the subsequent release of free amino acids into circulation. They also found an inverse relationship between elevated levels of Leu, Phe, Ala, and Lys in serum and decreased levels in tissues. This relationship is consistent with the observations made in our study, further reinforcing this perspective[ 24 , 25 ]. Additionally, Leu has been shown to stimulate 5’ adenosine monophosphate-activated protein kinase, a key regulator of cellular energy, thereby promoting favorable anabolic and catabolic processes in highly metabolically active tissues[ 26 ]. Phe, an essential amino acid, is a precursor of many proteins, and its flux can effectively represent systemic protein catabolism[ 27 ]. Therefore, we speculate that ectopic lesions induce alterations in metabolic status when they compromise normal tissues, and that the elevation of amino acids such as Leu and Phe in the serum constitutes a sensitive response to this process. Under certain conditions, Phe can be converted to Tyr by phenylalanine hydroxylase (PAH), and the ratio of Phe to Tyr is significantly elevated in patients with phenylketonuria due to PAH deficiency[ 28 ]. Moreover, other studies have shown that elevated Phe/Tyr ratios are strongly associated with immune activation and inflammation in patients with cardiovascular disease[ 29 ] and the progression of end-stage renal disease[ 30 ]. Our experiments revealed reduced Tyr levels in the EM group, suggesting that the disease state of EM may influence the metabolic processing of Phe in the body. Decreased Tyr levels are associated with EM and its carcinoid characteristics. Enhanced gluconeogenesis has been observed in heterogeneous cancers[ 31 ], indicating insufficient circulating glucose levels to support optimal cellular proliferation. Given that ectopic endometrial cells exhibit characteristics similar to those of cancer cells, gluconeogenesis may also increase in the EM to meet the increased energy demand of these cells. Our study revealed that Tyr, a ketogenic and gluconeogenic amino acid, is significantly decreased in the serum of patients with EM. This reduction may be attributed to the consumption of tyrosine, which fulfills the high-energy requirements of ectopic endometrial cells. Jana and Maignien also emphasized these metabolic alterations[ 32 , 33 ]. Endometriosis is a chronic inflammatory and immune disease characterized by increased levels of pro-inflammatory cytokines, growth factors, and angiogenesis, all contributing to disease persistence[ 34 ]. Our results revealed increased Kyn and decreased Trp levels in the EM group’s serum. Approximately 95% of ingested Trp enters the Kyn pathway, with indoleamine 2,3-dioxygenase 1 (IDO1) being the first enzyme stimulated by inflammatory molecules[ 35 ]. One study found that patients had a higher Kyn/Trp ratio than controls, indicating IDO1 activity[ 36 ]. In the present study, the Kyn/Trp ratio was elevated in the experimental group, inverting the presence of pro-inflammatory factors to stimulate IDO1 and confirming that EM is a chronic inflammatory and immune disease. Among the differentially expressed amino acids in our study, including H-Lys(Ac)-OH, 3-MeHis, and The, no prior reports document their elevation in the serum of patients with EM. H-Lys(Ac)-OH is an acetyl derivative of Lys. Lysine acetylation is a reversible modification regulated by two classes of enzymes: lysine acetyltransferases and lysine deacetylases. Acetylation targets nearly all cellular processes, thus linking this modification to cancer[ 37 , 38 ] and its progression in EM. Furthermore, H-Lys(Ac)-OH plays an essential role in glioblastoma development[ 39 ] and in patients with obesity and COVID-19[ 40 ]. 3-MeHis is formed by the posttranslational methylation of specific histidine residues in actin and myosin[ 41 ] and is released during the proteolysis of these proteins[ 42 ]. Myosin, a major component of thick filaments, is present in smooth and striated muscles. We hypothesized that an increase in 3-MeHis levels might indicate muscle loss when ectopic cells invade the myometrium. Theanine (N-ethyl-γ-glutamylamine), derived from tea leaves, plays an immunomodulatory role in inflammation, nerve injury, gastrointestinal health, and tumor progression. It regulates T lymphocyte function, glutathione synthesis, and cytokine and neurotransmitter levels[ 43 ]. Multiple studies have indicated that theanine has moderate apoptotic, antimetastatic, anti-migration, and anti-invasion effects on cancer cells. The mechanisms underlying the changes in the levels of these amino acids during EM progression remain unclear and warrant further investigation. Recent advances in metabolomics have led to the identification of numerous biomarkers associated with EM[ 15 ]. Many studies confirm that the predictive efficacy of multiple biomarkers significantly outperforms that of single ones, but biomarker selection and the establishment of diagnostic models vary widely. Unlike previous studies, this study developed five machine-learning models, each demonstrating superior performance. Machine learning is characterized by its ability to examine large amounts of data and discover correlations. This enhances diagnostic systems’ reliability, performance, and accuracy for various diseases. However, machine learning techniques are often called “black boxes” because they do not readily explain how predictions are generated, leading to mistrust among clinical users and hindering accurate judgment. To address this, we employed the SHAP method to interpret the “black box” of the machine learning models. The SHAP method provides global interpretations of the model’s overall function and ranks the feature values, clarifying the decision-making process [ 44 ]. Using this method to jointly select biomarkers for diagnostic model construction resulted in more comprehensive, efficient, and persuasive outcomes. Our study also leverages clinical data from subgroup analysis to provide a scientific basis for individualized medical decision-making. The ASRM score is based on the extent of the lesion, severity of symptoms (mainly dysmenorrhea), the size, and degree of adhesion. Among the EM samples collected, the lesions were located mainly in the ovary and myometrium, occurring as single or multiple. The size was calculated as the largest diameter. The ASRM scores of Class 3 differed from other subclasses and did not reflect lesion distribution, dysmenorrhea, or size, suggesting a more severe degree of adhesion. Clinically, the degree of adhesion between the lesion and the surrounding area profoundly affects the ease of EM surgery, choice of treatment, and postoperative recovery. Therefore, subclass analysis can help predict disease progression and provide ideas for managing surgery and individualized medical decision-making. However, our study had some limitations. First, the early EM diagnosis is challenging, and the inclusion criteria require laparoscopy, resulting in most patients being diagnosed with ASRM stages III/IV. Second, this was a single-center study with a limited sample size and representativeness, which may have introduced bias in the results. Multicenter studies or designs with larger sample sizes could enhance the generalizability and reliability of this study. Additionally, serum metabolic profile expression can be influenced by prior hormone treatment and the stages of the menstrual cycle during sampling[ 45 , 46 ]. Finally, numerous studies indicate that the recurrence rate of EM can reach 40–50% within 5 years post-surgery[ 47 ]. Consequently, iterations incorporating long-term follow-up studies to predict EM recurrence may demonstrate an even greater potency. 5. Conclusion This study employed metabolomic methods to identify new EM biological markers and used the SHAP method to increase the interpretability of machine learning algorithms. A non-invasive diagnostic scoring model that includes seven amino acids was successfully developed, demonstrating excellent predictive ability for EM in internal and external validations. Consequently, the model improves the efficiency and accuracy of EM diagnosis, monitors early disease progression and provides decision-support information for timely therapeutic intervention. This helps to protect patient fertility, increase cure rates, reduce healthcare costs and improve quality of life. In addition, the present study confirms that the development of EM is inextricably linked to its features (altered metabolic status, carcinoid features, chronic inflammatory diseases). These findings provide a mechanistic basis for the relevant aspects of amino acid metabolism in EM and guide the direction of future research. However, this study has the limitation of a single-center design, and further multicenter validation is needed to provide a more balanced view. Our next objective is to implement personalized treatment strategies based on the outcomes of the final predictive model, incorporate long-term follow-up studies, and predict future EM recurrence to improve the overall well-being of patients with EM. Declarations Ethics approval and consent to participate All human subjects gave written informed consent, and all human studies performed abided by the Declaration of Helsinki. The ethics committee of Women’s Hospital of Nanjing Medical University approved all study protocols (No. 2022KY-100). Funding statement This project was supported Nanjing Key Laboratory of Female Fertility Preservation and Restoration (No. 01002213), Nanjing Medical Science and Technique Development Foundation (Grant No. YKK24153), and Research Innovation Program for Graduates of Jiangsu Province (SJCX24_0755, SJCX23_0665). Author Contribution P.X., D.L. and Z.G. conceived the study, generated the hypotheses, and designed the experiments. M. L., S.X. and Y.C. performed the experiments and analyzed the data. S.X., F.L, N.Y., A.Y., T.F. J.C. and Y.G. carried out the clinical investigation and interpreted the clinical data. P.X., M.L. and D.L. drafted the manuscript. Acknowledgement N/A Data Availability All relevant data and materials are within this paper and its supplementary data. References Chinese Obstetricians and Gynecologists Association. Cooperative Group of Endometriosis, Chinese Society of Obstetrics and Gynecology, Chinese Medical Association. [Guideline for the diagnosis and treatment of endometriosis (Third edition)]. Zhonghua fu chan ke za zhi. 2021;56:812–24. Sonavane SK, Kantawala KP, Menias CO. Beyond the Boundaries—Endometriosis: Typical and Atypical Locations. Curr Probl Diagn Radiol. 2011;40:219–32. Aw H, Sa M. Pathophysiology, diagnosis, and management of endometriosis. BMJ (Clinical research ed) [Internet]. 2022 [cited 2025 Apr 11];379. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/36375827/ Practice Committee of the American Society for Reproductive Medicine. Endometriosis and infertility: a committee opinion. Fertil Steril. 2012;98:591–8. Kl PPPO, Endometriosis T. Epidemiology, Diagnosis and Clinical Management. Current obstetrics and gynecology reports [Internet]. 2017 [cited 2025 Apr 11];6. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29276652/ U LRM, A I SFGMAB. V G, A systematic review on endometriosis during pregnancy: diagnosis, misdiagnosis, complications and outcomes. Human reproduction update [Internet]. 2016 [cited 2025 Apr 11];22. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/26450609/ Cf Y, La MHPHZHKJ et al. D,. Improving the diagnosis of endometriosis in Asia-Pacific: Consensus from the Asia-Pacific Endometriosis Expert Panel for Endometriosis. International journal of gynaecology and obstetrics: the official organ of the International Federation of Gynaecology and Obstetrics [Internet]. 2023 [cited 2025 Apr 11];163. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37837343 /. R S, D S, A S, I P. An update on the biological markers of endometriosis. Minerva ginecologica [Internet]. 2017 [cited 2025 Apr 11];69. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/28271701/ Im YWKN. S. The Origin and Pathogenesis of Endometriosis. Annual review of pathology [Internet]. 2020 [cited 2025 Apr 11];15. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/31479615/ R G-R SD. L D. Nuclear factor-kappaB: a main regulator of inflammation and cell survival in endometriosis pathophysiology. Fertility and sterility [Internet]. 2012 [cited 2025 Apr 11];98. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/22771029/ Lc JV-JSH. G. The endometrial immune environment of women with endometriosis. Human reproduction update [Internet]. 2019 [cited 2025 Apr 11];25. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/31424502/ Lk S, Je M, Vr K, Rm M, K L. M K, The Immunopathophysiology of Endometriosis. Trends in molecular medicine [Internet]. 2018 [cited 2025 Apr 11];24. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/30054239/ G P, B B, M F, M R, A W. Environmental Factors and Endometriosis. International journal of environmental research and public health [Internet]. 2021 [cited 2025 Apr 11];18. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/34769544/ Zn L, Yf J, Jn R, Jh L. B D, J W. Amino acid metabolism in health and disease. Signal transduction and targeted therapy [Internet]. 2023 [cited 2025 Apr 11];8. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37699892/ Cn O, A T-R, Cb A. Metabolomics in endometriosis: challenges and perspectives for future studies. Reproduction & fertility [Internet]. 2021 [cited 2025 Apr 11];2. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/35128453/ G L, L W, G M. The pathophysiology of endometriosis and adenomyosis: tissue injury and repair. Archives of gynecology and obstetrics [Internet]. 2009 [cited 2025 Apr 11];280. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/19644696/ Ra SSXHEE, L B et al. Z, S K,. Complement pathway is frequently altered in endometriosis and endometriosis-associated ovarian cancer. Clinical cancer research: an official journal of the American Association for Cancer Research [Internet]. 2014 [cited 2025 Apr 11];20. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/25294912/ Li C, Xue J, Zhou D, Qi X, Chen T, Li Z. Integrating pharmacogenetics, multi-omics and machine learning in the novel therapeutic view of pulmonary hypertension. Interdisciplinary Med. 2025;3:e20240029. Bw M, Jg NB, Ma L, My W. B, F van der V, The performance of CA-125 measurement in the detection of endometriosis: a meta-analysis. Fertility and sterility [Internet]. 1998 [cited 2025 Apr 11];70. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/9848302/ Ke M, Sa C-H, Sh JVSK, Cm K. B. Peripheral biomarkers of endometriosis: a systematic review. Human reproduction update [Internet]. 2010 [cited 2025 Apr 11];16. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/20462942/ Jk RVTR, Er G. M. Endometriosis and the neoplastic process. Reproduction (Cambridge, England) [Internet]. 2004 [cited 2025 Apr 11];127. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/15016949/ Ms A, N P, Tm AA, Hm NMN. H, Cancer-Associated Mutations in Endometriosis without Cancer. The New England journal of medicine [Internet]. 2017 [cited 2025 Apr 11];376. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/28489996/ El L, T N. S R, J K. Amino acids in cancer. Experimental & molecular medicine [Internet]. 2020 [cited 2025 Apr 11];52. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/31980738/ E MDBSMJDD. S, M M, Metabolomics reveals perturbations in endometrium and serum of minimal and mild endometriosis. Scientific reports [Internet]. 2018 [cited 2025 Apr 11];8. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29691425/ J L, L G, H Z, Y G, J S, X G, et al. Endometrium metabolomic profiling reveals potential biomarkers for diagnosis of endometriosis at minimal-mild stages. Reproductive biology and endocrinology: RB&E [Internet]. 2018 [cited 2025 Apr 11];16. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29712562/ Np G, Ra V. Leucine–induced anabolic–catabolism: two sides of the same coin. Amino acids [Internet]. 2016 [cited 2025 Apr 11];48. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/26476924/ De M. An overview of phenylalanine and tyrosine kinetics in humans. The Journal of nutrition [Internet]. 2007 [cited 2025 Apr 11];137. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/17513423/ Jj M, Yj T, Cr S. Phenylalanine hydroxylase deficiency. Genetics in medicine: official journal of the American College of Medical Genetics [Internet]. 2011 [cited 2025 Apr 11];13. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/21555948/ C M, Tb G, Me AM. K, W M, D F. Immune activation and inflammation in patients with cardiovascular disease are associated with higher phenylalanine to tyrosine ratios: the ludwigshafen risk and cardiovascular health study. Journal of amino acids [Internet]. 2014 [cited 2025 Apr 11];2014. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/24660059/ Ks YBRAMB. N. Impairment of phenylalanine conversion to tyrosine in end-stage renal disease causing tyrosine deficiency. Kidney international [Internet]. 2004 [cited 2025 Apr 11];66. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/15253711/ Hj JC, X L, Sj WLH, L K. X, Gain of glucose-independent growth upon metastasis of breast cancer cells to the brain. Cancer research [Internet]. 2015 [cited 2025 Apr 11];75. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/25511375/ Sk J. M D, M J, S S, B C, K C. 1H NMR based targeted metabolite profiling for understanding the complex relationship connecting oxidative stress with endometriosis. BioMed research international [Internet]. 2013 [cited 2025 Apr 11];2013. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/23984345/ C M, C PSFK, K P-C CLM et al. Endometriosis phenotypes are associated with specific serum metabolic profiles determined by proton-nuclear magnetic resonance. Reproductive biomedicine online [Internet]. 2020 [cited 2025 Apr 11];41. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/32839101/ Ldgc R, Ms PSLM. A, F B, C C. Immunology of endometriosis. Best practice & research Clinical obstetrics & gynaecology [Internet]. 2018 [cited 2025 Apr 11];50. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29506962/ W Z, Ap S, K D, E G. S H, M K, Quantitative profiling of tryptophan metabolites in serum, urine, and cell culture supernatants by liquid chromatography-tandem mass spectrometry. Analytical and bioanalytical chemistry [Internet]. 2011 [cited 2025 Apr 11];401. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/21983980/ M LMRD, Ap OSM. L, M B, Functional expression of aryl hydrocarbon receptor on mast cells populating human endometriotic tissues. Laboratory investigation; a journal of technical methods and pathology [Internet]. 2016 [cited 2025 Apr 11];96. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/27348627/ M S, A A. Modulation of cellular processes by histone and non-histone protein acetylation. Nature reviews Molecular cell biology [Internet]. 2022 [cited 2025 Apr 11];23. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/35042977/ A R-T JG. S E-G. Lysine acetylation and cancer: A proteomics perspective. Journal of proteomics [Internet]. 2017 [cited 2025 Apr 11];150. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/27746255/ A BB, H L-N, Sc da SR. M S, E W, C V, Integrating Multi-Omics Analysis for Enhanced Diagnosis and Treatment of Glioblastoma: A Comprehensive Data-Driven Approach. Cancers [Internet]. 2023 [cited 2025 Apr 11];15. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37370767/ H A-H, N SS-A NJMH. A S, A E, N6-Acetyl-L-Lysine and p-Cresol as Key Metabolites in the Pathogenesis of COVID-19 in Obese Patients. Frontiers in immunology [Internet]. 2022 [cited 2025 Apr 11];13. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/35663953/ Jh ME, Wm C, Rs K. A. Complete amino-acid sequence of actin of rabbit skeletal muscle. Proceedings of the National Academy of Sciences of the United States of America [Internet]. 1973 [cited 2025 Apr 11];70. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/4517681/ Vr Y, Hn M. Ntau-methylhistidine (3-methylhistidine) and muscle protein turnover: an overview. Federation proceedings [Internet]. 1978 [cited 2025 Apr 11];37. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/350635/ K W SCJKHZ et al. Z H, L W,. L-Theanine and Immunity: A Review. Molecules (Basel, Switzerland) [Internet]. 2023 [cited 2025 Apr 11];28. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37175254/ Li Z. Extracting spatial effects from machine learning model using local interpretation method: An example of SHAP and XGBoost. Comput Environ Urban Syst. 2022;96:101845. Cf D, Ac KDBWAC. H, L B, Menstrual cycle rhythmicity: metabolic patterns in healthy women. Scientific reports [Internet]. 2018 [cited 2025 Apr 11];8. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/30275458/ Yz MW, Mj HMWMCFM. G, Effects of menstrual cycle phase on metabolomic profiles in premenopausal women. Human reproduction (Oxford, England) [Internet]. 2010 [cited 2025 Apr 11];25. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/20150174/ Sw G. Recurrence of endometriosis and its control. Human reproduction update [Internet]. 2009 [cited 2025 Apr 11];15. Available from: http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/19279046/ Additional Declarations No competing interests reported. Supplementary Files TableS1.docx TableS2.docx FIGURES100.tif Fig. S1. Differences in the abundance of amino acids. Abundance of 33 amino acids between the EM and NC groups. (ns, not significant; *, P -value < 0.05; **, P -value < 0.01; ***, P -value < 0.001). FIGURES200.tif Fig. S2. Machine-learning algorithms SHAP value ranking. SHAP values of each variable in the five machine-learning algorithms were displayed in descending order. The first row represented the training set, and the second row represented the test set. FIGURES300.tif Fig. S3. Total nomogram points and risk proportion. (A) Nomogram total points in the EM and NC groups (***, P -value < 0.001). (B) A confusion matrix was obtained based on the classification of high- and low-risk groups. FIGURES400.tif Fig. S4. Subclass analysis of EM group clinical characteristics. (A) The diameters in three subclasses. (B) The proportion of dysmenorrhea in three subclasses. (C) The proportion of single or multiple proportion in three subclasses. (ns, not significant). Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6696073","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":491541723,"identity":"9dff0939-f55b-4a78-bea1-f64b1b57596e","order_by":0,"name":"Moyuan Li","email":"","orcid":"","institution":"Nanjing Maternity and Child HealthCare Institute, Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Moyuan","middleName":"","lastName":"Li","suffix":""},{"id":491541724,"identity":"ed6192ab-0418-44b2-b5ca-5210a8612826","order_by":1,"name":"Sujuan Xu","email":"","orcid":"","institution":"Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Sujuan","middleName":"","lastName":"Xu","suffix":""},{"id":491541726,"identity":"e7790659-1808-4ebf-98d1-541de4f2907c","order_by":2,"name":"Yiwei Cao","email":"","orcid":"","institution":"The affiliated Jiangning Hospital of Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yiwei","middleName":"","lastName":"Cao","suffix":""},{"id":491541727,"identity":"913c7b17-2671-4ac0-9e1f-044ee0f4e0d1","order_by":3,"name":"Feiyang Li","email":"","orcid":"","institution":"Nanjing Maternity and Child HealthCare Institute, Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Feiyang","middleName":"","lastName":"Li","suffix":""},{"id":491541728,"identity":"4ce80c60-bdfc-46dc-b605-1dc2239e25aa","order_by":4,"name":"Nuo Ye","email":"","orcid":"","institution":"Nanjing Maternity and Child HealthCare Institute, Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Nuo","middleName":"","lastName":"Ye","suffix":""},{"id":491541729,"identity":"289a0284-f605-41c4-94d5-69be600af314","order_by":5,"name":"Yiran Xu","email":"","orcid":"","institution":"Nanjing Maternity and Child HealthCare Institute, Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Yiran","middleName":"","lastName":"Xu","suffix":""},{"id":491541730,"identity":"ecc05409-2554-4150-ae6e-8853d3897969","order_by":6,"name":"Jian Cao","email":"","orcid":"","institution":"Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Jian","middleName":"","lastName":"Cao","suffix":""},{"id":491541731,"identity":"fe9a4a5d-536b-4327-bf88-8f1405aab6d9","order_by":7,"name":"Aiyuan Yue","email":"","orcid":"","institution":"Nanjing Maternity and Child HealthCare Institute, Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Aiyuan","middleName":"","lastName":"Yue","suffix":""},{"id":491541733,"identity":"cc6b3d58-0d96-46b5-aa13-c66b0ad0327b","order_by":8,"name":"Tiantian Fan","email":"","orcid":"","institution":"Nanjing Maternity and Child HealthCare Institute, Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Tiantian","middleName":"","lastName":"Fan","suffix":""},{"id":491541734,"identity":"b87265e7-0af0-4988-ab8e-3984a028b0b9","order_by":9,"name":"Yichen Guo","email":"","orcid":"","institution":"Nanjing Maternity and Child HealthCare Institute, Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Yichen","middleName":"","lastName":"Guo","suffix":""},{"id":491541735,"identity":"af43c1a1-8479-4a72-9543-911a742adaf5","order_by":10,"name":"Zhen Gong","email":"","orcid":"","institution":"Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Zhen","middleName":"","lastName":"Gong","suffix":""},{"id":491541739,"identity":"617373f3-6f56-4912-a90c-c44f15582190","order_by":11,"name":"Dake Li","email":"","orcid":"","institution":"Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":false,"prefix":"","firstName":"Dake","middleName":"","lastName":"Li","suffix":""},{"id":491541740,"identity":"d18ffa3b-84ca-4369-b286-d606f48b34e2","order_by":12,"name":"Pengfei Xu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8UlEQVRIie3OsarCMBSA4VMK7RKcI0J9hUigKvgwKYJTFHG4OFyuBaGTOutb+Agpgi7BOcVB3RyuoLhfbEvn1LsJ5p/OgXzkAJhM79oYMIAbAsk38QqRGUHifyQNs2IrI013vrmIcStYJ1cxSiLwKopZj6GGtGf7XktITMlhwGg/AlpVzK4tNYQo7tN7hD1y4CQjwVoxx0Y6cvz1SfyHEUlkTiblRCF6isP0F4VywkgZac+4D2KLaVWmh/E9bqzkeVrTkaYr6U18/wSLnfQp/+rUK7tu/NAeBuDgYnbSJZutUANyYt+K2T5pn5pMJtPH9gTUU04Rj9SA0wAAAABJRU5ErkJggg==","orcid":"","institution":"Nanjing Maternity and Child HealthCare Institute, Women's Hospital of Nanjing medical University (Nanjing Women and Children's Healthcare Hospital)","correspondingAuthor":true,"prefix":"","firstName":"Pengfei","middleName":"","lastName":"Xu","suffix":""}],"badges":[],"createdAt":"2025-05-19 07:23:23","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6696073/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6696073/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87804042,"identity":"6097e8be-e32b-4ed8-9c2b-3ce0afc28f7c","added_by":"auto","created_at":"2025-07-29 08:12:59","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":864366,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWorkflow of this study. \u003c/strong\u003eFirst, quantification of serum amino acids from the NCs and EM patients was conducted by UPLC-MS/MS-based MRM method and variance analysis. Second, the abundance of serum amino acids was integrated with clinical features to establish predictive models using five machine-learning algorithms. The SHAP method selected potentially valuable variables, and a highly specific and sensitive diagnostic scoring panel was constructed. Finally, all EMs were differentiated into three subclasses based on the abundance of amino acid expression, and correlations among these classes, clinical manifestations, and prognosis were identified.\u003c/p\u003e","description":"","filename":"OnlineFIGURE1.png","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/6e7d298f3d7d67258ee3ce1e.png"},{"id":87804040,"identity":"6384a339-aabd-4517-83ab-41e74c2cd58c","added_by":"auto","created_at":"2025-07-29 08:12:59","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":179276,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEstablishment and validation of the amino acids quantification method. \u003c/strong\u003e(A) Retention time and ion pair information for 33 amino acids. (B) Standard curve of 33 amino acids: The linear relationship between concentration and intensity.\u003c/p\u003e","description":"","filename":"OnlineFIGURE200.png","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/109fac064eb7dc0ebd368fb6.png"},{"id":87804045,"identity":"6ed599ef-136e-4bf0-8efd-d014efdddb1a","added_by":"auto","created_at":"2025-07-29 08:12:59","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":191882,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentification of differentially abundant amino acids. \u003c/strong\u003e(A) PCA analysis between the EM and NC groups. (B) OPLS-DA among the EM and NC groups. (C) A heatmap for the abundance of amino acids with differences in the EM and NC groups. (D) A volcano plot illustrating differences between the EM and NC groups, with the color of the dots representing the P-value. (E) The area under the ROC curve for each amino acid.\u003c/p\u003e","description":"","filename":"OnlineFIGURE300.png","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/226d13f6241b96f58e3b462d.png"},{"id":87804048,"identity":"75afc57c-447a-48ab-8676-308f8e8e9ce1","added_by":"auto","created_at":"2025-07-29 08:12:59","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":115237,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePrediction model development and valuable variables selection. \u003c/strong\u003e(A) ROC curves for five machine-learning algorithms used to assess EM probability based on 33 amino acids and 13 clinical indicators. (B) PRC curves for five machine-learning algorithms used to assess EM probability based on 33 amino acids and 13 clinical indicators. (C) DCA curves for five machine-learning algorithms used to assess EM probability based on 33 amino acids and 13 clinical indicators. (D) Upset plot showed the intersection of the top 20 features across five machine-learning algorithms.\u003c/p\u003e","description":"","filename":"OnlineFIGURE400.png","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/0d74698a0a4a7d35fe8abfec.png"},{"id":87804055,"identity":"c615104a-d784-4769-86d5-aaea4bf1d948","added_by":"auto","created_at":"2025-07-29 08:12:59","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":99530,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConstructing adiagnostic scoring panel for predicting incident EM. \u003c/strong\u003e(A) A nomogram constructed from the 7 features to predict the probability of EM. (B) ROC curves for diagnostic scoring panel based on 7 amino acids. (C) PRC curves for diagnostic scoring panel based on 7 amino acids. (D) DCA curves for diagnostic scoring panel based on 7 amino acids. (E) Calibration curves for diagnostic scoring panel based on 7 amino acids. (F) Patients in training set were stratified into low-risk (\u0026lt;cutoff) and high-risk (\u0026gt;cutoff) groups based on the Youden index of the total point as the cutoff. Heatmap for the abundance of amino acids showed the differences. (G) Patients in test set were stratified into low-risk (\u0026lt;cutoff) and high-risk (\u0026gt;cutoff) groups based on the Youden index of the total point as the cutoff. Heatmap for the abundance of amino acids showed the differences.\u003c/p\u003e","description":"","filename":"OnlineFIGURE500.png","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/f945b97c88af7c9971703d06.png"},{"id":87804692,"identity":"6b5bf6b4-de5a-4c94-a429-de95f7ebf769","added_by":"auto","created_at":"2025-07-29 08:20:59","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":35341,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSubclass analysis and progression prediction. \u003c/strong\u003e(A) The UMAP clustering effect categorized the EM group into three subclasses based on amino acid abundance. (B) The proportion of stage Ⅲ and IV ASRM in three subclasses. (C) ASRM scores in three subclasses. (D) Nomogram total points in three subclasses. *, \u003cem\u003eP\u003c/em\u003e-value \u0026lt; 0.05; **, P-value \u0026lt; 0.01; ***, P-value \u0026lt; 0.001\u003c/p\u003e","description":"","filename":"OnlineFIGURE600.png","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/8dafb8fcf3f1c0feb581a9a6.png"},{"id":89400863,"identity":"96150df0-0005-446d-96c1-08fa07984bf6","added_by":"auto","created_at":"2025-08-19 14:17:16","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3123635,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/1ef0e034-9b30-4e4a-a619-8675d3ef93ac.pdf"},{"id":87804689,"identity":"95763506-1131-4485-88a8-386ef69dbe47","added_by":"auto","created_at":"2025-07-29 08:20:59","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":18516,"visible":true,"origin":"","legend":"","description":"","filename":"TableS1.docx","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/a8dfb63cbbc6f6312f16c9c5.docx"},{"id":87804044,"identity":"2e0005b8-123b-450c-b742-8a4af3f16eb6","added_by":"auto","created_at":"2025-07-29 08:12:59","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":18925,"visible":true,"origin":"","legend":"","description":"","filename":"TableS2.docx","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/f3d61f31526534054a353e27.docx"},{"id":87804051,"identity":"dbdf65c7-0902-4f77-8f90-8bdb3327ea96","added_by":"auto","created_at":"2025-07-29 08:12:59","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":12018926,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig. S1. Differences in the abundance of amino acids. \u003c/strong\u003eAbundance of 33 amino acids between the EM and NC groups. (ns, not significant; *, \u003cem\u003eP\u003c/em\u003e-value \u0026lt; 0.05; **, \u003cem\u003eP\u003c/em\u003e-value \u0026lt; 0.01; ***, \u003cem\u003eP\u003c/em\u003e-value \u0026lt; 0.001).\u003c/p\u003e","description":"","filename":"FIGURES100.tif","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/5365c24b16d26e53b1a6300e.tif"},{"id":87804694,"identity":"548fc548-901f-4713-bd2b-63d3c648104d","added_by":"auto","created_at":"2025-07-29 08:20:59","extension":"tif","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":20358658,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig. S2. Machine-learning algorithms SHAP value ranking. \u003c/strong\u003eSHAP values of each variable in the five machine-learning algorithms were displayed in descending order. The first row represented the training set, and the second row represented the test set.\u003c/p\u003e","description":"","filename":"FIGURES200.tif","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/304b8e8c71da9e571e909114.tif"},{"id":87804065,"identity":"5d212bcb-8b01-465b-bec3-2378844f0931","added_by":"auto","created_at":"2025-07-29 08:13:00","extension":"tif","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":18208258,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig. S3. Total nomogram points and risk proportion. \u003c/strong\u003e(A) Nomogram total points in the EM and NC groups (***, \u003cem\u003eP\u003c/em\u003e-value \u0026lt; 0.001). (B) A confusion matrix was obtained based on the classification of high- and low-risk groups.\u003c/p\u003e","description":"","filename":"FIGURES300.tif","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/ec23009a8c7668f3cc6ce0df.tif"},{"id":87804704,"identity":"287a6a56-b8b2-4f21-a056-2d9dd5f7b313","added_by":"auto","created_at":"2025-07-29 08:21:00","extension":"tif","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":18208258,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig. S4. Subclass analysis of EM group clinical characteristics. \u003c/strong\u003e(A) The diameters in three subclasses. (B) The proportion of dysmenorrhea in three subclasses. (C) The proportion of single or multiple proportion in three subclasses. (ns, not significant).\u003c/p\u003e","description":"","filename":"FIGURES400.tif","url":"https://assets-eu.researchsquare.com/files/rs-6696073/v1/b821603be6a7c59130491ea4.tif"}],"financialInterests":"No competing interests reported.","formattedTitle":"Interpretable machine-learning-derived diagnostic scoring panel for endometriosis identification: a study of serum amino acid profiling","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eEndometriosis (EM) is defined as the presence, growth, and infiltration of endometrial tissue (including glands and stroma) outside the uterine cavity, resulting in recurrent bleeding[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Ectopic endometrial glands undergo cyclic changes regulated by estrogen, leading to symptoms such as bleeding, pain, nodules or masses, and infertility, depending on the affected sites (including the ovaries, fallopian tubes, pelvic peritoneum, and pelvic ligaments)[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Previous studies indicate that 10\u0026ndash;15% of women of reproductive age and 70% of those with chronic pelvic pain have EM. Furthermore, approximately 25\u0026ndash;50% of infertile women have EM, and 30\u0026ndash;50% of those with EM experience infertility[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Despite this prevalence, diagnosis is often delayed by an average of 6.7 years among individuals aged 18\u0026ndash;45 years[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Such delays result in untimely treatment, adversely affecting fertility and increasing the risk of ovarian epithelial cancer, causing unnecessary suffering for women of reproductive age.\u003c/p\u003e\u003cp\u003eEndometriosis diagnosis has traditionally relied on symptoms and signs, ultrasound examinations, and laparoscopic surgery. Ultrasound examination results are highly dependent on the operator\u0026rsquo;s skill, while laparoscopic surgery, the gold standard, is invasive, high cost, potentially damaging, and carries anesthesia risks[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Although serum cancer antigen 125 is commonly used in clinical practice as a non-invasive diagnostic method, studies show its relatively inadequate sensitivity and specificity[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Consequently, developing a non-invasive diagnostic method for EM has become a high priority.\u003c/p\u003e\u003cp\u003eThe mechanisms underlying EM are intricate, encompassing theories such as retrograde menstruation[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], implantation of endometrial stem cells[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], inflammatory factors[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], the immune microenvironment[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], and others[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], which significantly affect the occurrence and progression. Amino acids are small molecules essential in manifold physiological activities in the human body, including protein synthesis, physiological regulation, immune function, and energy supply. Moreover, disorders in amino acid metabolism are closely associated with the occurrence and progression of multiple inflammatory diseases, including EM[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Other studies have revealed that amino acids are potential metabolomic biomarkers for EM diagnosis[\u003cspan additionalcitationids=\"CR16\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. However, the role of serum amino acid abundance combined with machine learning algorithms in the early diagnosis of EM remains limited.\u003c/p\u003e\u003cp\u003eMachine learning is the scientific study of algorithms and models that aim to enhance efficiency and accuracy in clinical diagnosis and assist physicians in decision-making[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The novel combination of serum amino acid profiling and machine learning has the potential to identify correlates of endometriosis, facilitating early, non-invasive diagnosis and enhancing patient outcomes. Herein, we characterized the serum amino acid profile in EM, screened for potentially valuable variables using assorted machine learning algorithms, and established a diagnostic scoring panel for predicting incident EM using a logistic regression-derived nomogram model consisting of seven amino acids with high specificity, sensitivity, and selectivity. This study provides new insights and a scientific basis for early EM diagnosis, mechanistic research, and subsequent development of personalized treatment and prevention strategies.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Preparation of Standards and Establishment of Standard Curves\u003c/h2\u003e\u003cp\u003eIn this study, 33 amino acids (3-methyl-L-histidine (3-MeHis), γ-aminobutyric acid (GABA), alanine (Ala), arginine (Arg), asparagine (Asn), aspartic acid (Asp), cystine (Cys), glutamic acid (Glu), glutamine (Gln), glycine (Gly), guanidinoacetic acid (Gua), histidine (His), homocitrulline (HCit), hydroxyproline (Hyp), isoleucine (Ile), kynurenine (Kyn), leucine (Leu), lysine (Lys), methionine (Met), N,N-dimethylglycine (Dmg), N6-acetyl-L-lysine (H-Lys(Ac)-OH), ornithine (Orn), phenylalanine (Phe), pipecolic acid (Pip), proline (Pro), pyroglutamic acid (PGA), S-adenosyl-L-homocysteine (SAH), serine (Ser), S-methyl-L-cysteine (SMC), theanine (The), threonine (Thr), tryptophan (Trp) and tyrosine (Tyr)) were purchased and prepared as a 5 mmol/L stock solution in 20% methanol aqueous solution, and stored at -80\u0026deg;C for future use. Precise aliquots of each standard stock mixture were stepwise diluted to prepare mixed standards for establishing the standard curve.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Multiple reaction monitoring (MRM) channel setup and Ultra-performance liquid chromatography-mass spectrometry (UPLC-MS/MS) conditions\u003c/h2\u003e\u003cp\u003eStandards for each substance were used to set the conditions for UPLC-MS/MS, including parent ion determination, parent ion fragmentation voltage optimization, daughter ion determination, daughter ion collision energy optimization, and MRM channel establishment. An Agilent 1260 liquid chromatography system coupled with a G6460 tertiary quadrupole mass spectrometer was used under the following chromatographic conditions: chromatographic column: Agilent Infinity Lab Poroshell 120 HIIC-Z (2.1\u0026times;100 mm, 2.7 \u0026micro;m); flow rate of 0.5 mL/min; column temperature of 25\u0026deg;C, and; mobile phase A: 20 mmol/L ammonium formate acetonitrile solution (V:water:V: acetonitrile\u0026thinsp;=\u0026thinsp;1:9); and mobile phase B: 20 mmol/L ammonium formate aqueous solution. Gradient elution was performed using the following linear gradient: A, 100% (0 min), 70% (11.5 min), 70% (12 min), and 100% (15 min). Mass spectrometry was performed in electrospray ionization positive-ion mode.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Patient recruitment and collection of clinical features\u003c/h2\u003e\u003cp\u003eBlood samples and clinical features were collected from 167 patients diagnosed with EM and 137 normal controls (NCs). These features include age, dysmenorrhea presence, blood test results conducted throughout the visit, including peripheral blood mononuclear cell (PBMC), cholesterol, triglycerides (TG), alanine aminotransferase (ALT), aspartate aminotransferase (AST), γ-glutamyl transpeptidase, high-density lipoprotein cholesterol, alpha-fetoprotein, carcinoembryonic antigen (CEA), cancer antigen 125 (CA 125), cancer antigen 199 (CA 199). A total of 167 patients with EM were diagnosed at Nanjing Medical University Affiliated Obstetrics and Gynecology Hospital between December 1, 2021, and December 1, 2023. All patients underwent laparoscopic examination with pathological confirmation of EM. The American Society for Reproductive Medicine (ASRM) stage, revised ASRM score, diameter, and whether the number of lesions were single or multiple were documented. Concurrently, 137 healthy female volunteers who underwent routine health checkups at our institution were recruited as the NC group. They had no chronic diseases, medication histories, or abnormalities on pelvic examination. The exclusion criteria were as follows: (1) pregnant or breastfeeding women; (2) aged\u0026thinsp;\u0026lt;\u0026thinsp;18 years or \u0026gt;\u0026thinsp;50 years; (3) history of medication treatment within the past 3 months; (4) other comorbidities such as hyperthyroidism, hypothyroidism, diabetes, or hypertension; and (5) refusal to participate or were uncooperative with the researchers. All participants provided written informed consent, and the research adhered to the Declaration of Helsinki. The ethics committee of the Nanjing Medical University Affiliated Gynecology Hospital approved all research protocols (No. 2022 KY-100).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Serum amino acids extraction\u003c/h2\u003e\u003cp\u003eBlood samples from the EM group were collected before surgery, while those from the NC group were collected after \u0026gt;\u0026thinsp;8 h of overnight fasting. The serum from blood samples was centrifuged again to obtain the supernatant, which was then transferred to an Eppendorf tube. Subsequently, 800 \u0026micro;L of pre-cooled methanol (containing 12.5 \u0026micro;mol/L of L-glutamate 13C5 solution) was added, and the mixture was vortexed vigorously for 5 min. Subsequently, the samples were incubated at -80\u0026deg;C for 8 h to ensure complete protein precipitation. Following incubation, the samples were centrifuged at 16,000 \u0026times; g at 4\u0026deg;C for 15 min. The supernatant was carefully transferred to new tubes and lyophilized at 4\u0026deg;C until powdered. Before loading, the sample was re-dissolved in 100 \u0026micro;L of a solution containing 80% methanol and 20% water (methanol: water\u0026thinsp;=\u0026thinsp;4:1) and further centrifuged for visualization. The supernatant was then transferred to a mass spectrometry vial.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Data processing\u003c/h2\u003e\u003cp\u003eDifferences in clinical baseline characteristics were analyzed using the chi-square test for categorical data and the Student\u0026rsquo;s t-test (two-tailed) or one-way analysis of variance for continuous data. Agilent MassHunter Qualitative Analysis B.06.00 software was used to extract and analyze retention times, peak areas, and peak heights of 33 amino acids, normalizing data to an internal standard. Raw data underwent principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) using R, evidencing reproducibility within groups and differences between groups, thereby establishing a model correlating amino acid expression levels with group classification. Independent samples were compared using the Student\u0026rsquo;s t-test (two-tailed). In the heatmap and volcano plot analyses, filtered significantly up- and downregulated amino acids based on p\u0026thinsp;\u0026lt;\u0026thinsp;0.05 and a fold change (FC)\u0026thinsp;\u0026gt;\u0026thinsp;1.5 or \u0026lt;\u0026thinsp;0.67.\u003c/p\u003e\u003cp\u003eThe samples were subsequently categorized into training and testing sets at a 7:3 ratio. Thirty-three amino acids and 13 clinical indicators, including random forest (RF), neural network (NN), extreme gradient boosting, support vector classification (SVC), and naive Bayes, were used to establish multiple machine learning algorithms using Python software. The predictive performance of these models was evaluated using receiver operating characteristic (ROC) curves, precision-recall curves (PRC), and decision curve analysis (DCA). Given the clinical challenge of accepting predictions from models that lack interpretability, the SHapley Additive exPlanations (SHAP) method was utilized for the global interpretation of the machine learning algorithm results, highlighting the importance of features and selecting potentially valuable variables for constructing a diagnostic scoring panel. The selected biomarkers were analyzed using logistic regression analysis in R, and a nomogram was designed to represent disease risk prediction visually. The predictive performances of the five machine learning algorithms and the diagnostic scoring panel were evaluated using ROC curves, PRC, and DCA. The diagnostic score for each subject and total nomogram points were calculated from the nomogram model, and the ROC curve for total nomogram points determined the optimal threshold, classifying individuals with scores above the cutoff as high risk for EM.\u003c/p\u003e\u003cp\u003eWe characterized the EM group into several subclasses in terms of amino acid abundance using an unsupervised learning algorithm, K-Means Clustering. The uniform manifold approximation and projection (UMAP) algorithm subsequently demonstrated the efficacy of the clustering effect. Differences in the ASRM scores, diameters, and nomogram total points between subclasses were analyzed using the Wilcoxon test, while ASRM stage, single or multiple lesions, and percentage of dysmenorrhea were analyzed using Fisher\u0026rsquo;s exact test.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Workflow\u003c/h2\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the study workflow. First, metabolites were extracted from the samples using a liquid-liquid extraction process, followed by quantification of serum amino acids using the UPLC-MS/MS-based MRM method and variance analysis. Second, the abundance of serum amino acids was integrated with clinical features to establish predictive models using five machine-learning algorithms. The SHAP method selected potentially valuable variables, and a highly specific and sensitive diagnostic scoring panel was constructed before evaluating its effectiveness. Finally, all EMs were differentiated into three subclasses based on the abundance of amino acid expression, and correlations among these classes, clinical manifestations, and prognosis were identified.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Clinical characteristics of the participants\u003c/h2\u003e\u003cp\u003eThis study included a cohort of 304 participants: 167 individuals diagnosed with EM and 137 NCs. The detailed demographic and clinical characteristics of all subjects are presented in \u003cb\u003eTable S1\u003c/b\u003e. Overall, differences in the number of peripheral PBMCs and TG, ALT, AST, and CEA levels were observed between the two groups. However, the levels of CA 125 and CA 199 in the EM group were significantly higher than those in the NC group. Additionally, patients with EM were more likely to experience dysmenorrhea than NC participants. No statistically significant differences in other demographic or clinical indicators were observed between the EM and NC groups (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). In the EM group, 84 single lesions were observed in the ovarian endometriosis (OvE) and 22 in the adenomyosis. Among multiple lesions, 48 were found in the OvE and 13 in the OvE combined with adenomyosis.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Establishment and validation of the amino acid\u0026rsquo;s quantification method\u003c/h2\u003e\u003cp\u003eWe performed UPLC-MS/MS-based serum amino acid profiling. MS/MS spectrometric data, including parent and fragment ion information for each amino acid, were selected, and optimization of the decluttering potential and collision energy facilitated the establishment of the MRM method, with chromatograms of 33 amino acids shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA. The retention time for each metabolite was determined using a mixed standard solution of amino acids. Linear gradient elution effectively separated 33 amino acids within 10 min, with peak times primarily concentrated between 2 and 8 min. Excellent linearity was observed across various amino acid concentration ranges (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). \u003cb\u003eTable S2\u003c/b\u003e showed the regression equation, r2 and linear range for 33 amino acids.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e3.4 Identification of differentially abundant amino acids\u003c/h2\u003e\u003cp\u003eUsing the established UPLC-MS/MS method, the retention times, peak areas, and peak heights of the amino acids were extracted and normalized, yielding corresponding levels for each sample. The PCA and OPLS-DA results revealed a separation trend between the EM and NC groups, with Q\u0026sup2; = 0.735 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA and \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). Statistical analysis revealed a significant difference in the abundance of 28 amino acids (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) (\u003cb\u003eFig. S1\u003c/b\u003e). Amino acids exhibiting FC\u0026thinsp;\u0026gt;\u0026thinsp;1.5 or \u0026lt;\u0026thinsp;0.67 and a \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05 were considered differentially abundant. The abundances of 3-MeHis, Kyn, Leu, H-Lys(Ac)-OH, Phe, The, Tyr, etc., were greater in EM samples than in the NC samples, except for Tyr, which was lower. Hierarchical cluster analysis revealed intragroup correlations and significant intergroup differences (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC). These amino acids are visible in the volcano plot (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD). The abundance of Kyn, Phe, Leu, and Tyr consistently differed between the two groups. Furthermore, analysis of the area under the ROC curve for each amino acid indicated that most had the potential to identify EM (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE). A trend of smaller P-values correlating with larger areas under curve values was observed, suggesting that differences in amino acid abundance enhanced predictive performance.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e3.5 Prediction model development and valuable variables selection\u003c/h2\u003e\u003cp\u003eThis study incorporated 33 amino acids and 13 clinical indicators as features to generate five machine-learning algorithms for assessing the risk of developing EM. The predictive performances of these five models in the training and test sets are shown in Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eA-C. The ROC curve analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eA) and PRC analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eB) indicated that the area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC) for all machine learning models in the training set was greater than 0.9, while in the testing set, they were both greater than 0.7. In the DCA, the net benefit curves of all machine learning models exhibited increased net benefit within a certain threshold probability range, coupled with a concomitant smooth transition (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eC).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eGenerally, the SVC and NN models showed greater net benefits, although the predictive performances of the other models are noteworthy and should not be overlooked.\u003c/p\u003e\u003cp\u003eThe SHAP method was employed to select valuable variables by calculating the contribution of each variable. As shown in \u003cb\u003eFigure S2\u003c/b\u003e, the average SHAP values were used to assess the contribution of the variables to each model, which are displayed in descending order. For example, it ranked first in its contribution to the overall function of the RF model in training and testing sets. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e4\u003c/span\u003eD, the intersection of the top 20 features across models revealed seven common features: 3-MeHis, Kyn, Leu, H-Lys(Ac)-OH, Phe, The, and Tyr.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e3.6 Constructing a diagnostic scoring panel for predicting incident EM\u003c/h2\u003e\u003cp\u003eThrough global interpretation using SHAP values, we identified the top seven common features as valuable variables for early EM diagnosis. A diagnostic scoring panel for predicting incident EM was constructed using a logistic regression nomogram (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e5\u003c/span\u003eA). This panel yielded AUROC values of 0.96 and 0.89 for the training and testing sets, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e5\u003c/span\u003eB), coupled with AUPRC values of 0.94 and 0.87, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e5\u003c/span\u003eC). The DCA results revealed satisfactory potential clinical benefits (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e5\u003c/span\u003eD). These findings emphasize the excellent predictive performance of the diagnostic scoring panel. Furthermore, 1,000 bootstrap resampling iterations with replacement were conducted, and the calibration curve demonstrated the robustness of the diagnostic scoring panel (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e5\u003c/span\u003eE). These findings emphasize the high predictive performance of the diagnostic scoring panel. Finally, each feature\u0026rsquo;s score was summed to obtain a total nomogram point, with a cutoff value of 159.86 determined by the Youden index, allowing for the classification of high-risk (nomogram total point\u0026thinsp;\u0026gt;\u0026thinsp;cutoff value) and low-risk (nomogram total point\u0026thinsp;\u0026lt;\u0026thinsp;cutoff value) groups (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e5\u003c/span\u003eF and \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e5\u003c/span\u003eG). The heatmap revealed that the abundance of most amino acids, except tyrosine, was greater in the high-risk group, suggesting that tyrosine was a low-risk feature while others were high-risk. The total nomogram points in the EM group were significantly greater than those in the NC group (\u003cb\u003eFig. S3A\u003c/b\u003e), and this difference was statistically significant. A confusion matrix was obtained based on the classification of high- and low-risk groups. In the training set, the true-positive rate for predicting incident EM was 115/116 (99.14%), and the true-negative rate was 86/96 (89.58%). In the testing set, the true positive rate for predicting incident EM was 49/51 (96.08%), and the true negative rate was 31/41 (75.61%) (\u003cb\u003eFig. S3B\u003c/b\u003e). These results strongly indicate that this diagnostic scoring panel has significant value and clinical applicability for early diagnosis of EM.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003e3.7 Subclass analysis and progression prediction\u003c/h2\u003e\u003cp\u003eThe UMAP clustering effect (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e6\u003c/span\u003eA) categorized the EM group into three subclasses based on amino acid abundance. The proportion of patients with stage IV ASRM (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e6\u003c/span\u003eB) and ASRM scores (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e6\u003c/span\u003eC) was significantly greater in class 3 than in the other two classes (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), indicating more extensive lesions and severe symptoms. From the diagnostic scoring panel\u0026rsquo;s perspective, Class 3 patients had higher total nomogram points (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e6\u003c/span\u003eD, P\u0026thinsp;\u0026lt;\u0026thinsp;0.05) and a greater risk of developing EM. Additionally, the diameter (\u003cb\u003eFig. S4A\u003c/b\u003e), presence of dysmenorrhea (\u003cb\u003eFig. S4B\u003c/b\u003e), and single or multiple proportions (\u003cb\u003eFig. S4C\u003c/b\u003e) did not differ among the three subclasses. Based on these results, we can predict disease progression using subclass analysis and provide ideas for managing surgery and personalized treatment.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThis study used UPLC-MS/MS to determine the abundance of amino acids in the serum samples of patients with EM. In the variance analysis, 11 amino acids were significantly increased, whereas Tyr levels decreased in the EM group. Five machine-learning algorithms were subsequently developed using the SHAP method for potentially valuable variables. Finally, a diagnostic scoring panel comprising seven amino acids (3-MeHis, Kyn, Leu, H-Lys(Ac)-OH, Phe, The, and Tyr) was established using a logistic regression-derived nomogram model, demonstrating excellent potential for early EM diagnosis. This analysis identified specific amino acids that could serve as biomarkers for the early diagnosis of EM.\u003c/p\u003e\u003cp\u003eCurrently, non-invasive and potentially valuable variables for the early diagnosis of EM are still lacking. Patients with EM often experience symptoms such as bleeding, dysmenorrhea, abdominal masses, and infertility[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. This condition severely affects women of childbearing age in China. However, diagnosis is frequently delayed, which affects prognosis and increases the risk of malignancy. However, laparoscopic and ultrasonographic examinations have limitations. Consequently, research efforts focus on developing non-invasive detection methods, such as liquid testing (blood, urine, and follicular fluid), to improve early diagnosis accuracy. CA125 is the most studied and used biomarker for EM, but its sensitivity and specificity are inadequate[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], showing greater efficacy in diagnosing advanced (stage III-IV) than early (stage I-II) endometriosis[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. To date, over 100 potential EM biomarkers have been reported, but few are applicable to mild EM detection[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Therefore, applying machine-learning algorithms to identify reliable and consistent biomarkers is crucial. This approach established an early EM diagnostic model, helping to reduce infertility, cancer risk, and complications associated with late-stage diagnosis.\u003c/p\u003e\u003cp\u003eDisruptions in amino acid metabolism are closely associated with EM progression, metabolic disorders, immune disorders, and cancer. Ectopic endometrial cells exhibit many similarities to cancer cells, including uncontrolled proliferation, cell invasion, neovascularization, and apoptosis resistance, as is widely recognized in the literature[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Therefore, it is hypothesized that the changes in amino acids during the progression of EM and cancer may be parallel. Traditionally, cancer metabolism has focused on central carbon metabolism, including glycolysis and the tricarboxylic acid cycle. However, recent studies have revealed that amino acids also play fundamental roles in energy regulation, biosynthetic support, and redox balance[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Similar to tumors, alterations in amino acid metabolism can serve as clinical indicators of EM progression and inform treatment strategies.\u003c/p\u003e\u003cp\u003eOur study detected a significant increase in the levels of several amino acids, notably Leu, Phe, Ala, and Lys, in the serum of patients with EM compared to normal NCs. These changes may be attributed to physiological mechanisms related to tissue damage and biosynthetic repair associated with EM. In support of this notion, Dutta et al. suggested that the catabolic state induced by tissue damage in EM leads to an increased breakdown of endogenous proteins and the subsequent release of free amino acids into circulation. They also found an inverse relationship between elevated levels of Leu, Phe, Ala, and Lys in serum and decreased levels in tissues. This relationship is consistent with the observations made in our study, further reinforcing this perspective[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Additionally, Leu has been shown to stimulate 5\u0026rsquo; adenosine monophosphate-activated protein kinase, a key regulator of cellular energy, thereby promoting favorable anabolic and catabolic processes in highly metabolically active tissues[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Phe, an essential amino acid, is a precursor of many proteins, and its flux can effectively represent systemic protein catabolism[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Therefore, we speculate that ectopic lesions induce alterations in metabolic status when they compromise normal tissues, and that the elevation of amino acids such as Leu and Phe in the serum constitutes a sensitive response to this process. Under certain conditions, Phe can be converted to Tyr by phenylalanine hydroxylase (PAH), and the ratio of Phe to Tyr is significantly elevated in patients with phenylketonuria due to PAH deficiency[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Moreover, other studies have shown that elevated Phe/Tyr ratios are strongly associated with immune activation and inflammation in patients with cardiovascular disease[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] and the progression of end-stage renal disease[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Our experiments revealed reduced Tyr levels in the EM group, suggesting that the disease state of EM may influence the metabolic processing of Phe in the body.\u003c/p\u003e\u003cp\u003eDecreased Tyr levels are associated with EM and its carcinoid characteristics. Enhanced gluconeogenesis has been observed in heterogeneous cancers[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], indicating insufficient circulating glucose levels to support optimal cellular proliferation. Given that ectopic endometrial cells exhibit characteristics similar to those of cancer cells, gluconeogenesis may also increase in the EM to meet the increased energy demand of these cells. Our study revealed that Tyr, a ketogenic and gluconeogenic amino acid, is significantly decreased in the serum of patients with EM. This reduction may be attributed to the consumption of tyrosine, which fulfills the high-energy requirements of ectopic endometrial cells. Jana and Maignien also emphasized these metabolic alterations[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eEndometriosis is a chronic inflammatory and immune disease characterized by increased levels of pro-inflammatory cytokines, growth factors, and angiogenesis, all contributing to disease persistence[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Our results revealed increased Kyn and decreased Trp levels in the EM group\u0026rsquo;s serum. Approximately 95% of ingested Trp enters the Kyn pathway, with indoleamine 2,3-dioxygenase 1 (IDO1) being the first enzyme stimulated by inflammatory molecules[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. One study found that patients had a higher Kyn/Trp ratio than controls, indicating IDO1 activity[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. In the present study, the Kyn/Trp ratio was elevated in the experimental group, inverting the presence of pro-inflammatory factors to stimulate IDO1 and confirming that EM is a chronic inflammatory and immune disease.\u003c/p\u003e\u003cp\u003eAmong the differentially expressed amino acids in our study, including H-Lys(Ac)-OH, 3-MeHis, and The, no prior reports document their elevation in the serum of patients with EM. H-Lys(Ac)-OH is an acetyl derivative of Lys. Lysine acetylation is a reversible modification regulated by two classes of enzymes: lysine acetyltransferases and lysine deacetylases. Acetylation targets nearly all cellular processes, thus linking this modification to cancer[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e] and its progression in EM. Furthermore, H-Lys(Ac)-OH plays an essential role in glioblastoma development[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e] and in patients with obesity and COVID-19[\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. 3-MeHis is formed by the posttranslational methylation of specific histidine residues in actin and myosin[\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e] and is released during the proteolysis of these proteins[\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Myosin, a major component of thick filaments, is present in smooth and striated muscles. We hypothesized that an increase in 3-MeHis levels might indicate muscle loss when ectopic cells invade the myometrium. Theanine (N-ethyl-γ-glutamylamine), derived from tea leaves, plays an immunomodulatory role in inflammation, nerve injury, gastrointestinal health, and tumor progression. It regulates T lymphocyte function, glutathione synthesis, and cytokine and neurotransmitter levels[\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. Multiple studies have indicated that theanine has moderate apoptotic, antimetastatic, anti-migration, and anti-invasion effects on cancer cells. The mechanisms underlying the changes in the levels of these amino acids during EM progression remain unclear and warrant further investigation.\u003c/p\u003e\u003cp\u003eRecent advances in metabolomics have led to the identification of numerous biomarkers associated with EM[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Many studies confirm that the predictive efficacy of multiple biomarkers significantly outperforms that of single ones, but biomarker selection and the establishment of diagnostic models vary widely. Unlike previous studies, this study developed five machine-learning models, each demonstrating superior performance. Machine learning is characterized by its ability to examine large amounts of data and discover correlations. This enhances diagnostic systems\u0026rsquo; reliability, performance, and accuracy for various diseases. However, machine learning techniques are often called \u0026ldquo;black boxes\u0026rdquo; because they do not readily explain how predictions are generated, leading to mistrust among clinical users and hindering accurate judgment. To address this, we employed the SHAP method to interpret the \u0026ldquo;black box\u0026rdquo; of the machine learning models. The SHAP method provides global interpretations of the model\u0026rsquo;s overall function and ranks the feature values, clarifying the decision-making process [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Using this method to jointly select biomarkers for diagnostic model construction resulted in more comprehensive, efficient, and persuasive outcomes.\u003c/p\u003e\u003cp\u003eOur study also leverages clinical data from subgroup analysis to provide a scientific basis for individualized medical decision-making. The ASRM score is based on the extent of the lesion, severity of symptoms (mainly dysmenorrhea), the size, and degree of adhesion. Among the EM samples collected, the lesions were located mainly in the ovary and myometrium, occurring as single or multiple. The size was calculated as the largest diameter. The ASRM scores of Class 3 differed from other subclasses and did not reflect lesion distribution, dysmenorrhea, or size, suggesting a more severe degree of adhesion. Clinically, the degree of adhesion between the lesion and the surrounding area profoundly affects the ease of EM surgery, choice of treatment, and postoperative recovery. Therefore, subclass analysis can help predict disease progression and provide ideas for managing surgery and individualized medical decision-making.\u003c/p\u003e\u003cp\u003eHowever, our study had some limitations. First, the early EM diagnosis is challenging, and the inclusion criteria require laparoscopy, resulting in most patients being diagnosed with ASRM stages III/IV. Second, this was a single-center study with a limited sample size and representativeness, which may have introduced bias in the results. Multicenter studies or designs with larger sample sizes could enhance the generalizability and reliability of this study. Additionally, serum metabolic profile expression can be influenced by prior hormone treatment and the stages of the menstrual cycle during sampling[\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. Finally, numerous studies indicate that the recurrence rate of EM can reach 40\u0026ndash;50% within 5 years post-surgery[\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]. Consequently, iterations incorporating long-term follow-up studies to predict EM recurrence may demonstrate an even greater potency.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study employed metabolomic methods to identify new EM biological markers and used the SHAP method to increase the interpretability of machine learning algorithms. A non-invasive diagnostic scoring model that includes seven amino acids was successfully developed, demonstrating excellent predictive ability for EM in internal and external validations. Consequently, the model improves the efficiency and accuracy of EM diagnosis, monitors early disease progression and provides decision-support information for timely therapeutic intervention. This helps to protect patient fertility, increase cure rates, reduce healthcare costs and improve quality of life. In addition, the present study confirms that the development of EM is inextricably linked to its features (altered metabolic status, carcinoid features, chronic inflammatory diseases). These findings provide a mechanistic basis for the relevant aspects of amino acid metabolism in EM and guide the direction of future research. However, this study has the limitation of a single-center design, and further multicenter validation is needed to provide a more balanced view. Our next objective is to implement personalized treatment strategies based on the outcomes of the final predictive model, incorporate long-term follow-up studies, and predict future EM recurrence to improve the overall well-being of patients with EM.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e\u003cp\u003e All human subjects gave written informed consent, and all human studies performed abided by the Declaration of Helsinki. The ethics committee of Women\u0026rsquo;s Hospital of Nanjing Medical University approved all study protocols (No. 2022KY-100).\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding statement\u003c/h2\u003e\u003cp\u003eThis project was supported Nanjing Key Laboratory of Female Fertility Preservation and Restoration (No. 01002213), Nanjing Medical Science and Technique Development Foundation (Grant No. YKK24153), and Research Innovation Program for Graduates of Jiangsu Province (SJCX24_0755, SJCX23_0665).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eP.X., D.L. and Z.G. conceived the study, generated the hypotheses, and designed the experiments. M. L., S.X. and Y.C. performed the experiments and analyzed the data. S.X., F.L, N.Y., A.Y., T.F. J.C. and Y.G. carried out the clinical investigation and interpreted the clinical data. P.X., M.L. and D.L. drafted the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eN/A\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll relevant data and materials are within this paper and its supplementary data.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eChinese Obstetricians and Gynecologists Association. Cooperative Group of Endometriosis, Chinese Society of Obstetrics and Gynecology, Chinese Medical Association. [Guideline for the diagnosis and treatment of endometriosis (Third edition)]. Zhonghua fu chan ke za zhi. 2021;56:812\u0026ndash;24.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSonavane SK, Kantawala KP, Menias CO. Beyond the Boundaries\u0026mdash;Endometriosis: Typical and Atypical Locations. Curr Probl Diagn Radiol. 2011;40:219\u0026ndash;32.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAw H, Sa M. Pathophysiology, diagnosis, and management of endometriosis. BMJ (Clinical research ed) [Internet]. 2022 [cited 2025 Apr 11];379. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/36375827/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/36375827/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePractice Committee of the American Society for Reproductive Medicine. Endometriosis and infertility: a committee opinion. Fertil Steril. 2012;98:591\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKl PPPO, Endometriosis T. Epidemiology, Diagnosis and Clinical Management. Current obstetrics and gynecology reports [Internet]. 2017 [cited 2025 Apr 11];6. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29276652/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29276652/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eU LRM, A I SFGMAB. V G, A systematic review on endometriosis during pregnancy: diagnosis, misdiagnosis, complications and outcomes. Human reproduction update [Internet]. 2016 [cited 2025 Apr 11];22. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/26450609/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/26450609/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCf Y, La MHPHZHKJ et al. D,. Improving the diagnosis of endometriosis in Asia-Pacific: Consensus from the Asia-Pacific Endometriosis Expert Panel for Endometriosis. International journal of gynaecology and obstetrics: the official organ of the International Federation of Gynaecology and Obstetrics [Internet]. 2023 [cited 2025 Apr 11];163. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37837343\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37837343\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e/.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eR S, D S, A S, I P. An update on the biological markers of endometriosis. Minerva ginecologica [Internet]. 2017 [cited 2025 Apr 11];69. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/28271701/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/28271701/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIm YWKN. S. The Origin and Pathogenesis of Endometriosis. Annual review of pathology [Internet]. 2020 [cited 2025 Apr 11];15. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/31479615/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/31479615/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eR G-R SD. L D. Nuclear factor-kappaB: a main regulator of inflammation and cell survival in endometriosis pathophysiology. Fertility and sterility [Internet]. 2012 [cited 2025 Apr 11];98. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/22771029/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/22771029/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLc JV-JSH. G. The endometrial immune environment of women with endometriosis. Human reproduction update [Internet]. 2019 [cited 2025 Apr 11];25. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/31424502/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/31424502/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLk S, Je M, Vr K, Rm M, K L. M K, The Immunopathophysiology of Endometriosis. Trends in molecular medicine [Internet]. 2018 [cited 2025 Apr 11];24. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/30054239/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/30054239/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eG P, B B, M F, M R, A W. Environmental Factors and Endometriosis. International journal of environmental research and public health [Internet]. 2021 [cited 2025 Apr 11];18. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/34769544/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/34769544/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZn L, Yf J, Jn R, Jh L. B D, J W. Amino acid metabolism in health and disease. Signal transduction and targeted therapy [Internet]. 2023 [cited 2025 Apr 11];8. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37699892/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37699892/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCn O, A T-R, Cb A. Metabolomics in endometriosis: challenges and perspectives for future studies. Reproduction \u0026amp; fertility [Internet]. 2021 [cited 2025 Apr 11];2. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/35128453/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/35128453/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eG L, L W, G M. The pathophysiology of endometriosis and adenomyosis: tissue injury and repair. Archives of gynecology and obstetrics [Internet]. 2009 [cited 2025 Apr 11];280. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/19644696/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/19644696/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRa SSXHEE, L B et al. Z, S K,. Complement pathway is frequently altered in endometriosis and endometriosis-associated ovarian cancer. Clinical cancer research: an official journal of the American Association for Cancer Research [Internet]. 2014 [cited 2025 Apr 11];20. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/25294912/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/25294912/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi C, Xue J, Zhou D, Qi X, Chen T, Li Z. Integrating pharmacogenetics, multi-omics and machine learning in the novel therapeutic view of pulmonary hypertension. Interdisciplinary Med. 2025;3:e20240029.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBw M, Jg NB, Ma L, My W. B, F van der V, The performance of CA-125 measurement in the detection of endometriosis: a meta-analysis. Fertility and sterility [Internet]. 1998 [cited 2025 Apr 11];70. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/9848302/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/9848302/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKe M, Sa C-H, Sh JVSK, Cm K. B. Peripheral biomarkers of endometriosis: a systematic review. Human reproduction update [Internet]. 2010 [cited 2025 Apr 11];16. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/20462942/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/20462942/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJk RVTR, Er G. M. Endometriosis and the neoplastic process. Reproduction (Cambridge, England) [Internet]. 2004 [cited 2025 Apr 11];127. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/15016949/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/15016949/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMs A, N P, Tm AA, Hm NMN. H, Cancer-Associated Mutations in Endometriosis without Cancer. The New England journal of medicine [Internet]. 2017 [cited 2025 Apr 11];376. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/28489996/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/28489996/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEl L, T N. S R, J K. Amino acids in cancer. Experimental \u0026amp; molecular medicine [Internet]. 2020 [cited 2025 Apr 11];52. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/31980738/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/31980738/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eE MDBSMJDD. S, M M, Metabolomics reveals perturbations in endometrium and serum of minimal and mild endometriosis. Scientific reports [Internet]. 2018 [cited 2025 Apr 11];8. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29691425/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29691425/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJ L, L G, H Z, Y G, J S, X G, et al. Endometrium metabolomic profiling reveals potential biomarkers for diagnosis of endometriosis at minimal-mild stages. Reproductive biology and endocrinology: RB\u0026amp;E [Internet]. 2018 [cited 2025 Apr 11];16. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29712562/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29712562/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNp G, Ra V. Leucine\u0026ndash;induced anabolic\u0026ndash;catabolism: two sides of the same coin. Amino acids [Internet]. 2016 [cited 2025 Apr 11];48. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/26476924/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/26476924/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDe M. An overview of phenylalanine and tyrosine kinetics in humans. The Journal of nutrition [Internet]. 2007 [cited 2025 Apr 11];137. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/17513423/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/17513423/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJj M, Yj T, Cr S. Phenylalanine hydroxylase deficiency. Genetics in medicine: official journal of the American College of Medical Genetics [Internet]. 2011 [cited 2025 Apr 11];13. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/21555948/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/21555948/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eC M, Tb G, Me AM. K, W M, D F. Immune activation and inflammation in patients with cardiovascular disease are associated with higher phenylalanine to tyrosine ratios: the ludwigshafen risk and cardiovascular health study. Journal of amino acids [Internet]. 2014 [cited 2025 Apr 11];2014. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/24660059/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/24660059/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKs YBRAMB. N. Impairment of phenylalanine conversion to tyrosine in end-stage renal disease causing tyrosine deficiency. Kidney international [Internet]. 2004 [cited 2025 Apr 11];66. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/15253711/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/15253711/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHj JC, X L, Sj WLH, L K. X, Gain of glucose-independent growth upon metastasis of breast cancer cells to the brain. Cancer research [Internet]. 2015 [cited 2025 Apr 11];75. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/25511375/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/25511375/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSk J. M D, M J, S S, B C, K C. 1H NMR based targeted metabolite profiling for understanding the complex relationship connecting oxidative stress with endometriosis. BioMed research international [Internet]. 2013 [cited 2025 Apr 11];2013. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/23984345/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/23984345/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eC M, C PSFK, K P-C CLM et al. Endometriosis phenotypes are associated with specific serum metabolic profiles determined by proton-nuclear magnetic resonance. Reproductive biomedicine online [Internet]. 2020 [cited 2025 Apr 11];41. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/32839101/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/32839101/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLdgc R, Ms PSLM. A, F B, C C. Immunology of endometriosis. Best practice \u0026amp; research Clinical obstetrics \u0026amp; gynaecology [Internet]. 2018 [cited 2025 Apr 11];50. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29506962/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/29506962/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eW Z, Ap S, K D, E G. S H, M K, Quantitative profiling of tryptophan metabolites in serum, urine, and cell culture supernatants by liquid chromatography-tandem mass spectrometry. Analytical and bioanalytical chemistry [Internet]. 2011 [cited 2025 Apr 11];401. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/21983980/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/21983980/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eM LMRD, Ap OSM. L, M B, Functional expression of aryl hydrocarbon receptor on mast cells populating human endometriotic tissues. Laboratory investigation; a journal of technical methods and pathology [Internet]. 2016 [cited 2025 Apr 11];96. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/27348627/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/27348627/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eM S, A A. Modulation of cellular processes by histone and non-histone protein acetylation. Nature reviews Molecular cell biology [Internet]. 2022 [cited 2025 Apr 11];23. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/35042977/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/35042977/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eA R-T JG. S E-G. Lysine acetylation and cancer: A proteomics perspective. Journal of proteomics [Internet]. 2017 [cited 2025 Apr 11];150. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/27746255/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/27746255/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eA BB, H L-N, Sc da SR. M S, E W, C V, Integrating Multi-Omics Analysis for Enhanced Diagnosis and Treatment of Glioblastoma: A Comprehensive Data-Driven Approach. Cancers [Internet]. 2023 [cited 2025 Apr 11];15. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37370767/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37370767/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eH A-H, N SS-A NJMH. A S, A E, N6-Acetyl-L-Lysine and p-Cresol as Key Metabolites in the Pathogenesis of COVID-19 in Obese Patients. Frontiers in immunology [Internet]. 2022 [cited 2025 Apr 11];13. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/35663953/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/35663953/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJh ME, Wm C, Rs K. A. Complete amino-acid sequence of actin of rabbit skeletal muscle. Proceedings of the National Academy of Sciences of the United States of America [Internet]. 1973 [cited 2025 Apr 11];70. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/4517681/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/4517681/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVr Y, Hn M. Ntau-methylhistidine (3-methylhistidine) and muscle protein turnover: an overview. Federation proceedings [Internet]. 1978 [cited 2025 Apr 11];37. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/350635/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/350635/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eK W SCJKHZ et al. Z H, L W,. L-Theanine and Immunity: A Review. Molecules (Basel, Switzerland) [Internet]. 2023 [cited 2025 Apr 11];28. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37175254/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/37175254/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi Z. Extracting spatial effects from machine learning model using local interpretation method: An example of SHAP and XGBoost. Comput Environ Urban Syst. 2022;96:101845.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCf D, Ac KDBWAC. H, L B, Menstrual cycle rhythmicity: metabolic patterns in healthy women. Scientific reports [Internet]. 2018 [cited 2025 Apr 11];8. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/30275458/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/30275458/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYz MW, Mj HMWMCFM. G, Effects of menstrual cycle phase on metabolomic profiles in premenopausal women. Human reproduction (Oxford, England) [Internet]. 2010 [cited 2025 Apr 11];25. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/20150174/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/20150174/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSw G. Recurrence of endometriosis and its control. Human reproduction update [Internet]. 2009 [cited 2025 Apr 11];15. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/19279046/\u003c/span\u003e\u003cspan address=\"http://pubmed-ncbi-nlm-nih-gov-s.webvpn.njmu.edu.cn:8118/19279046/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Endometriosis, serum amino acids, diagnostic scoring panel, machine learning, SHAP","lastPublishedDoi":"10.21203/rs.3.rs-6696073/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6696073/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDelayed diagnosis of endometriosis (EM) can significantly hinder treatment options and patient prognosis, adversely affecting fertility and increasing the risk of malignancy. Current diagnostic methods are invasive. In our study, ultra-high-performance liquid chromatography-tandem mass spectrometry-based serum amino acid profiling was used to compare amino acid abundance between 137 normal controls (NCs) and 167 patients with EM. Subsequently, five machine learning algorithms (random forest, neural network, extreme gradient boosting, support vector classification, and na\u0026iuml;ve Bayes) were used to establish the efficacy and stability of amino acid for diagnosing. Ultimately, seven amino acids, 3-methyl-L-histidine, kynurenine, leucine, N6-acetyl-L-lysine, phenylalanine, theanine, and tyrosine, were successfully identified as valuable variables using the SHapley Additive exPlanations (SHAP) method and were included in the diagnostic scoring panel. The areas under the receiver operating characteristic and precision-recall curves of the diagnostic scoring panel were greater than 0.8. Additionally, we characterized the EM group into several subclasses using K-Means clustering and analyzed differences between the subclasses. In conclusion, This study developed a non-invasive diagnostic scoring panel with excellent predictive ability. Consequently, it enhances the accuracy of EM diagnosis, mitigates the risk of malignancy, and facilitates the formulation of subsequent personalized treatment and prevention strategies.\u003c/p\u003e","manuscriptTitle":"Interpretable machine-learning-derived diagnostic scoring panel for endometriosis identification: a study of serum amino acid profiling","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-29 08:12:54","doi":"10.21203/rs.3.rs-6696073/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b1b35625-9dfd-40a2-a18d-e0abd66504b8","owner":[],"postedDate":"July 29th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-08-19T14:09:02+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-29 08:12:54","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6696073","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6696073","identity":"rs-6696073","version":["v1"]},"buildId":"B-jG_2CBjPDmsCi4Wdhf-","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}