| Objective: Endometrial carcinoma is one of the three malignant tumors of reproductive system that seriously threaten women’s health.There is a huge difference in the survival of patients with cancers at early and advanced stage.The lack of sensitive and specific biomarkers is the bottleneck of non-invasive early diagnosis of endometrial cancer.It is of great clinical significance to explore biomarkers with high sensitivity and specificity.Endometrial cancer is a recognized metabolic disease,and the abnormality of metabolic markers plays an important role in the early prediction of the disease and the anchoring of therapeutic targets.Due to the abundance of blood metabolites and the availability of clinical samples,blood metabolites have become the best samples for biomarker discovery,the research of which is also considered as an important application field of metabonomics.Lipid metabolism is an important energy source of tumor.Lipid,as an important signal transduction molecule,can not only affect cell proliferation,cycle and apoptosis,but also regulate the interaction between cells.Meanwhile,lipid is closely related to the evolution,progression,metastasis and other processes of tumor.As an important intermediate in lipid metabolism,free fatty acids have been confirmed to be abnormally expressed in multiple cancers,and are expected to become metabolic markers for cancer screening and early diagnosis.It is recognized that endometrial adenocarcinoma is closely associated with abnormal lipid metabolism.Therefore,by studying the plasma free fatty acid profile of endometrial cancer,it is expected to find ideal biomarkers for screening and early diagnosis.However,since a single metabolite in blood varies greatly due to diet,underlying diseases,mood and other factors,and the organism in the disease state is not only the change of one or several metabolic markers,but the change of the whole metabolic pattern.Therefore,the application of high-throughput mass spectrometry detection method combined with big data modeling method can effectively solve the influence of interference factors and achieve the purpose of accurate diagnosis.Common medical modeling methods include classical statistical logistic regression and artificial intelligence technology.The former requires data to conform to linear assumptions and is not as accurate as the latter for interpreting complex,multidimensional and nonlinear relationships in biological systems.Artificial intelligence technology has made important breakthroughs in the medical field by integrating and reprocessing disease information.Commonly used artificial intelligence algorithms include single algorithm,integrated algorithm and neural network algorithm.Each algorithm has its own advantages and disadvantages,and it is difficult to choose in the transformation application.In recent years,with the in-depth research of various artificial intelligence models in the field of medicine,the research on the comparison between the models has gradually attracted attention.However,the construction of metabolic molecular diagnostic model of endometrial cancer is rarely reported.This new field is an urgent clinical problem to be solved at present,and it is a breakthrough to solve the lack of early screening and early diagnosis of endometrial cancer.Therefore,this study aims to establish a molecular diagnostic model for endometrial cancer metabolism in the context of free fatty acid profile data.Methods: The plasma samples of 363 patients recruited from February 2020 to August2022 were targeted for 23 kinds of free fatty acids by liquid chromatogram-mass spectrometry method with optimized multi-reaction monitoring technology.The mass spectrometry data were processed by Analyst MD software,and the contents of each fatty acid in plasma were further quantified by standard curve method.Finally,the quantitative results of 23 free fatty acids were determined.In the correlation analysis between the quantitative results of 23 kinds of free fatty acids and clinicopathology,the Mann-Whitney U test was used to analyze the experimental group and the control group,and the median and interquartile range were used to describe the statistics and screen the differential metabolites.Further univariate analysis and multivariate analysis were performed to identify the independent risk factors associated with disease progression.The differential metabolites were used as dependent variables to construct a Logistic regression diagnostic model and external verification was performed.Furthermore,by using the Scikit-Learn framework package in Python software,the differential metabolites are taken as features,the data is preprocessed by the standardization method,the features are selected by the variance filtering method and the embedding method,and the parameters of the model were adjusted by the combination of cross-validation and learning curve.Five models including decision tree,random forest,support vector machine,Lasso regression and XGBoost were constructed,and the diagnostic efficacy of the five models in the diagnosis of endometrial cancer was compared horizontally to select and verify the model externally.Results: 1.Liquid chromatography-mass spectrometry was used to quantitatively detect23 kinds of free fatty acids in 363 plasma samples.Among them,the quantitative results of plasma free fatty acids in samples of 130 patients with endometrial adenocarcinoma and 93 patients in control group were statistically compared.12 abnormal free fatty acids,including myristic acid,palmitic acid,arachidic acid,myristoleic acid,palmitoleic acid,oleic acid,eicosadienoic acid,mead acid,linoleic acid,linolenic acid,DPA,and DTA,had significant statistical difference.The expression of endometrial carcinoma group was raised(P < 0.05).Further analysis of the clinical data of 12 different metabolites showed that mead acid was significantly associated with deep muscle infiltration(P=0.013),while eicosadienoic acid was significantly associated with tumor diameter(P=0.029).In addition,12 differential metabolites were used as diagnostic markers to diagnose endometrial cancer,and the area under the receiver operating characteristic curve(AUC)of the 12 indicators was the highest(AUC=0.704).The Logistic model of combined diagnosis was constructed using 12 differential metabolites.The sensitivity of the training test set was 73.8%,the specificity was 65.6%,the accuracy was 70.4%,and the AUC was 0.74.The sensitivity of the validation set was 61.4%,the specificity was 84.2%,the accuracy was 72.9%,and the AUC was 0.74.The positive predictive value was80.8%,and the negative predictive value was 68.2%,which proved that the diagnostic performance of multi-target combined diagnostic model was better than that of single metabolite.2.Firstly,we preprocessed the quantitative results of the 12 differential metabolites screened in the first part,used the standardization method to conduct dimensionless processing on the data,and applied the variance filtering method and embedding method for preliminary screening to complete the feature selection.The parameters of each prediction model were adjusted by combining 10-fold cross-validation and learning curve,and the optimal parameters were used to construct the five models.The accuracy of the training test set of decision tree,random forest,Lasso regression,support vector machine and XGBoost model were 70%,74%,65%,74% and 87%,respectively.The sensitivity of each model was 69%,92%,71%,86%,92%;the specificity was 70%,55%,56%,56%,80%;and the AUC was 0.7,0.73,0.63,0.71,0.86.Further external verification of the built XGBoost model showed the accuracy,specificity,sensitivity,AUC,positive predictive value and negative predictive value of validation set were 80.0%,71.4%,88.6%,0.80,75.6% and 86.2%,respectively.Finally,the diagnostic performance of XGBoost and Logistic regression model was significantly improved.Conclusions: 1.23 kinds of free fatty acids were determined by liquid chromatography-mass spectrometry.12 abnormal free fatty acids,including myristic acid,palmitic acid,arachidic acid,myristoleic acid,palmitoleic acid,oleic acid,eicosadienoic acid,mead Acid,linoleic acid,linolenic acid,DPA,and DTA,expressed abnormally high in plasma of patients with endometrial adenocarcinoma.Mead acid was found to be associated with muscular invasion,while eicosadienoic acid was significantly associated with tumor diameter.The accuracy of multi-target Logistic diagnostic model combined with 12 differential metabolites in the diagnosis of endometrial cancer was higher than that of single metabolite.2.Compared with decision tree,random forest,support vector machine and Lasso regression,the XGBoost model constructed with 12 differential metabolites was more accurate in the diagnosis of endometrial cancer,and the diagnostic performance was significantly higher than that of Logistic diagnostic model. |
PDF Full Text Request