Fuzzy Data Mining and Bioinformatics Analysis in Methylation Analysis of M6A Gene Promoter Region in Esophageal Cancer

Wu, Shuoming; Yang, Xiangbao; Qiu, Xie; Pan, Yinpeng

doi:https://doi.org/10.1155/2022/4420717

Journal of Sensors

On this page

Abstract Introduction Materials and Methods Results Discussion Conclusions Data Availability Conflicts of Interest References Copyright Related Articles

Special Issue

Advanced Machine Learning and Big Data Analytics with IoT Sensor Data

View this Special Issue

Research Article | Open Access

Volume 2022 | Article ID 4420717 | https://doi.org/10.1155/2022/4420717

Fuzzy Data Mining and Bioinformatics Analysis in Methylation Analysis of M6A Gene Promoter Region in Esophageal Cancer

Shuoming Wu,¹Xiangbao Yang,¹Xie Qiu,¹and Yinpeng Pan¹

Academic Editor: Sweta Bhattacharya

Received28 Jul 2022

Accepted18 Aug 2022

Published07 Sept 2022

Abstract

This work was aimed at analyzing the correlation between the methylation level of the M6A gene in esophageal cancer (EC) and the prognosis of patients based on bioinformatics technology and evaluating the prognostic predictive values of different data mining models. 80 EC patients and 80 healthy people were selected, and the serum of the patients was collected to detect the level of DNA methyltransferase. During the radical resection of EC, tumor tissues and adjacent normal tissues were collected from patients to detect the methylation level of the M6A gene. COX regression analysis was employed to analyze the independent risk factors (IRFs) of M6A gene methylation and other treatments affecting the prognosis of EC patients. The particle swarm optimization (PSO) algorithm was introduced to improve the fuzzy -means clustering (FCM) algorithm. The differences in the prognostic prediction efficiency of logistic regression analysis (LRA), decision tree (DT) C5.0, artificial neural network (ANN), support vector machine (SVM), and improved FCM (IFCM) models were compared. The levels of DNA methyltransferase and human histone deacetylase 1 (HSD-1) in EC patients were increased greatly (). The methylation rates and methylation levels of M6A methylation regulators (ALKBH5, HNRNPC, METTL3, WTAP, RBM15, YTHDC1, YTHDF1, and FTO) in EC tissues were obviously higher (). The survival time of high-risk EC patients was much shorter than that of low-risk patients (). Univariate and multivariate COX regression analysis showed that gender, tumor grade, TNM grade, degree of infiltration, and methylation of ALKBH5, HNRNPC, and METTL3 genes were IRFs for the prognosis of EC patients (). The areas under the ROC curve (AUCs) of LRA, DT C5.0, ANN, SVM, and IFCM algorithms for predicting the prognosis of patients were 0.813, 0.857, 0.895, 0.926, and 0.958, respectively, and the IFCM model had the best diagnostic effect. In conclusion, the detection of bioinformatics technology showed no obvious DNA methylation in EC patients, and the elevated levels of M6A methylation regulators in patients were an IRF affecting the prognosis of patients. In addition, the fuzzy data mining model can be undertaken as the preferred method for prognosis prediction of EC patients.

1. Introduction

Esophageal cancer (EC) is a malignant tumor of the upper gastrointestinal tract that occurs in the esophageal epithelial tissue. Histologically, it can be classified into esophageal squamous cell carcinoma and esophageal adenocarcinoma [1]. EC is one of the six most common malignant tumors, and its incidence and pathological types have obvious regional differences [2]. The EC morbidity and mortality of residents in high-risk areas are 500 times higher than those in endemic areas. Moreover, the prognosis of EC patients is poor, and the 5-year survival rate is only about 10%, and there is no obvious improvement after the improvement of medical level [3]. Due to the high morbidity and mortality of EC, there is an urgent need to find better biomarkers, which can facilitate the early diagnosis of EC and improve the prognosis of patients. Current studies suggest that genetic changes and epigenetic changes are the main mechanisms of tumor-induced tumor at the gene level [4]. Genetic changes refer to changes in the DNA base sequence, such as gene mutation, microsatellite instability, and deletion of heterozygosity. Epigenetic changes refer to the chemical modification of DNA itself and the regulation of DNA expression at the transcriptional level [5]. Epigenetics includes DNA methylation, genomic stress, and chromatin remodeling. Epigenetics is very common during the occurrence of malignant tumors, mainly the methylation of CpG islands in the promoter regions of tumor suppressor genes [6]. DNA methylation markers can be used in the classification and detection of diseases and can also reflect early changes in the occurrence and development of cancer. Therefore, exploring the DNA methylation of tumor suppressor genes is of great significance for the early diagnosis, risk assessment, prognosis prediction, and monitoring of EC.

RNA methylation modification accounts for 60% of all RNA modifications, and N6-methyladenodine (M6A) is the most common modification on mRNA and ncRNAs in higher organisms [7]. M6A is a methylation modification that occurs on RNA adenine (A). Current studies have found that miRNA, circRNA, tRNA, and snoRNA all have M6A modifications, which are mainly regulated by M6A methylation regulators [8]. Studies have confirmed that M6A methylation is closely related to the development and progression of tumors, and the expression level of M6A regulatory factors may directly determine the pathological process of malignant tumors [9]. Currently known M6A methyltransferase complexes include METTL3 and METTL14, and ALKBH5 and FTO can be considered as methylases to reverse methylation [10]. M6A-binding protein contains YTH domain proteins (YTHDF1, YTHDF2, YTHDC1, YTHDC2, etc.) and nuclear heterogeneity protein HNRNP family (HNRNPC, etc.) [11]. Abnormally expressed M6A methylation regulators can lead to abnormal M6A methylation modification, which in turn affects RNA processing, mRNA degradation, and translation and ultimately affects gene expression and promotes tumorigenesis and development [12]. Therefore, M6A methylation regulators can be used as potential prognostic biomarkers for the early diagnosis of EC and the prognosis prediction of patients.

In this work, bioinformatics methods and COX regression analysis were employed to find EC-related biomarkers, and a risk model for prognosis prediction of patients was constructed to explore the role of M6A methylation regulators in EC prognosis evaluation. Subsequently, an improved fuzzy data mining model was constructed, and the accuracy of different data mining models for prognosis prediction of EC patients was compared, aiming to provide a reference for improving the early diagnosis rate and prognosis prediction accuracy of EC.

2. Materials and Methods

2.1. Research Objects

80 patients who were diagnosed with EC from March 2020 to January 2022 during clinicopathological examinations in The First People’s Hospital of Lianyungang were selected, including 47 males and 33 females, with a median age of 63 years. The 80 normal controls were from healthy people who underwent physical examination in the medical examination department of The First People’s Hospital of Lianyungang during the same period, including 44 males and 36 females, with a median age of 60 years. The inclusion criteria were given as follows: those who met the clinical diagnostic criteria for esophageal cancer and were diagnosed by pathology, patients with complete clinical and follow-up data; and those who had not received chemotherapy or radiotherapy. The exclusion criteria were given as follows: patients with incomplete clinical and follow-up data and healthy people with no malignant tumors. This experiment had been approved by the Clinical Ethics Committee of the The First People’s Hospital of Lianyungang, and all patients signed informed consent.

2.2. Determination of Serum Methylation-Related Protein Expression Levels by Enzyme-Linked Immunosorbent Assay (ELISA)

2 mL of fasting venous peripheral blood samples was taken from EC patients and healthy people to take the supernatant after centrifugation. Serum DNA methyltransferases (DNMTs) and human histone deacetylase 1 (HDAC1) were detected by ELISA. DNMTs include DNMT1, DNMT3a, and DNMT3b. The DNMTs and HDAC1 antibodies were coated with specific antibodies, respectively, and solid-phase antibodies were prepared. Then, the solid-phase carrier of the corresponding protein was added to the solid-phase antibody microwells and incubated. Then, horseradish peroxidase-labeled DNMTs and HDAC1 antibody were added to form antibody complexes and washed. 3,3,5,5-Tetramethylbenzidine was added for color development, and the absorbance of each well was measured at a wavelength of 450 nm using a microplate reader to calculate the DNMTs and HDAC1 protein concentrations.

2.3. Surgical Treatment

All EC patients were treated with EC radical mastectomy. Routine skin preparation was performed before surgery, and artificial pneumothorax was established after tracheal anesthesia. A thoracoscopic observation hole was constructed in the seventh intercostal space of the patient’s right axilla to determine the location of the lesion. Three operating holes were established in the third intercostal space of the anterior axillary line and the sixth intercostal space of the posterior axillary line, and surgical-related instruments were inserted into the thoracic tissue, and the mediastinal pleura was incised. The lymph nodes in the tracheoesophageal groove and the surrounding loose tissue were removed, and the esophagus was dissected, and then, lymph node dissection was performed. Subsequently, the surrounding involved tissues were excised, followed by hemostasis, indwelling drainage tube, sputum suction, and closure of pneumothorax. Finally, the patient’s cancer tissue and adjacent normal tissue (normal esophageal epithelial tissue at 5 cm around the cancer tissue) were collected for follow-up experiments.

2.4. Determination of DNA Gene Methylation

The tissue DNA extraction kit was adopted to extract DNA from EC tissue, and a UV spectrophotometer was applied to detect the concentration and purity of extracted DNA, and the DNA that met the requirements is stored in a -20°C refrigerator for later use. The DNA CpG Island Methylation Modification Kit was employed for bisulfite modification of the extracted DNA, and the bisulfite modified DNA was purified according to the instructions of the Promega Wizard Clean-up DNA Purification and Recovery Kit. Sequence numbers of 10 M6A genes including ALKBH5, HNRNPC, METTL3, METTL14, WTAP, RBM15, YTHDC1, YTHDF1, YTHDF2, and FTO were searched from the GenBank database (https://www.ncbi.nlm.nih.gov/gene/). The primers, including methylated and unmethylated primer sequences, were quantitatively detected by Shanghai Sangon Bioengineering Co., Ltd. The reaction system of PCR amplification was 1.5 mmol/L MgCl₂, 0.2 mmol/L dNTPs, 200 nmol/L primer, and 2 μL bisulfite modified DNA template, and ddH₂O was used to make up to 25 μL. The degree of PCR reaction was set as 95°C for 10 min, adding 1.0 U hot-start polymerase; 95°C for 30 s, 58°C/56°C for 30 s, 72°C for 30 s (35 cycles); and 72°C for 5 min. In addition, it should prepare 2% agarose gel electrophoresis to detect the methylation level of each gene.

2.5. Data Mining Model

In this work, the input data including patient age, gender, tumor differentiation, TNM stage, tumor infiltration, M6A gene methylation, and other data were very neat, without missing values and in line with normal distribution, so no data transformation was required. For data mining, the sigmoid function in the model required the normalization of the original data to be in the range of [0,1]. Therefore, the maximum and minimum method was employed to normalize the data, and the processing equation was given as follows:

In the above equation (1), was the minimum value in the entire column of data, and referred to the maximum value in the entire column of data.

The sampling function of SPSS 19.0 was adopted, and the data were randomly divided into a training set and a test set with a ratio of 3 : 1. The training set included 59 normal controls and 61 EC patients, and the test set included 21 normal controls and 19 EC patients. Based on the sorted data, LRA, DT C5.0, ANN, SVM, and FCM algorithms were employed to construct the lung cancer-normal control prediction model. (1)The LRA model was a generalized linear regression analysis model. When it was used in this work, the main parameters in the model needed to be set as multinomial LRA, the modeling method was entered, the basic classification target was 1.0, and the model type was the main effect. The LRA model was trained by using the training set(2)The DT C5.0 algorithm was an improvement of the C4.5 model, which can be used for the analysis of data sets containing large amounts of data [13]. Compared with the C4.5 model, the C5.0 model showed the advantages of fast speed, small scale of generating DT, high classification accuracy, and many options. The information gain of the model was given as follows:

In equation (2) above, represented the data set, and represented a certain value. The main parameters in the DT C5.0 model used in this work were set to partition selection and output DT. Accuracy was improved by bootstrapping with a pruning severity of 75 and a minimum number of records for subbranch 2. It should ensure global pruning, no discriminative attributes, and no error classification loss. (3)ANN is a group of connected units or nodes called artificial neurons, which is a loose modeling method of neurons in biological brains. Each link had a weight, which determined the strength of the node’s influence on another node. At present, the most common ANN model is the back propagation (BP) network [14]. Therefore, the training parameters of the ANN model constructed in this work are set to be selected by the quick method. The number of random seeds was 10, the training setting was 50%, the cycle time was 1 min, the optimization condition was memory, the number of hidden layers was 1, and the number of hidden layer nodes was 15. The number of continuations was 200, the initial learning rate was 0.05, the minimum value was 0.01, the maximum value was 0.10, and the decay amount was 25(4)SVM is a two-class classification model, which is defined as a linear classifier with the largest interval in the feature space [15]. Because the interval was too large, the SVM model was different from the perceptron. The main parameters when constructing the SVM model in this work were set as nonlinear, , kernel type as polynomial, and (5)FCM algorithm is a widely used fuzzy data mining algorithm, but it is very easy to fall into the local extreme value, and the algorithm is sensitive to the setting of the initial value [16]. For this reason, the PSO algorithm was adopted in this work to improve and optimize the FCM algorithm into IFCM. The speed update equation of the standard PSO algorithm was as follows:

In equation (3) above, was the inertia weight value. The larger the was, the particles would perform a global search with a larger step size, and the local development would be finer. In the iterative process of the algorithm, decreased linearly, which can be expressed as below equation:

In equation (4), was the maximum number of iterations, referred to the current number of iterations, was the maximum inertia weight, and represented the minimum inertia weight.

The PSO algorithm was adopted to improve the FCM algorithm, and the calculation steps of the algorithm are shown in Figure 1. Firstly, the PSO algorithm was used to solve the global optimal clustering center: (1) the initial setting of the clustering division of the FCM algorithm was carried out. It should set the initial position of the particle and the optimal position of the individual in the PSO algorithm and initialize the particle velocity to obtain the global optimal position and fitness. The initial number of iterations was . (2) The position and velocity of the individual in the particle swarm and the inertia weight were updated. (3) It should set the cluster center according to the position of the particle and calculate the attribute value of the particle and the membership matrix of the FCM. (4) It should determine the individual optimal value of the particle and the overall optimal value. (5) Judgment of the termination condition was performed. If satisfied, it could output the global optimal solution and enter the next stage of calculation; if not, it should return to step (2). (6) It should take the global optimal solution as the initial clustering center, initialize the parameters of the FCM algorithm, and calculate the membership matrix. (7) It could calculate the cluster center in the FCM algorithm after iteration and calculate the membership function. (8) The condition for determining whether the threshold of iteration was full or not. If satisfied, the clustering had converged and the clustering result can be terminated; if not, it should return to step (7).

The data mining models LRA, DT C5.0, ANN, and SVM adopted in this work needed to be constructed by SPSS 19.0. The IFCM algorithm needed to be implemented using Matlab R2010a software. Before processing the experimental data, it was also necessary to use the data set in the UCI machine learning database to test the algorithm. The data set used in this work to test the clustering performance of the IFCM algorithm was Glass, which contained 214 research objects, the number of clusters was 6, and the attribute was 9.

2.6. Statistical Analysis

SPSS 19.0 was employed to organize and analyze the data. Enumeration data were expressed by (%), and differences were compared using the chi-square test. Measurement data were expressed as , and differences were compared using test. Kaplan-Meier curves were drawn for EC patient survival analysis. The factors affecting the prognosis of EC patients were analyzed by Cox regression analysis model, the hypothesis test of regression coefficients was set as the likelihood ratio test, and variable screening was set as backward. The diagnostic tests were used to evaluate the prediction effects of various data mining models. Evaluation indicators included sensitivity (Se), specificity (Sp), accuracy (Acc), AUC, positive predictive value (PPV), and negative predictive value (NPV). was the best. The test level was (two-tailed), and was considered statistically significant.

3. Results

3.1. General Data

The differences between the general data of EC patients and normal controls were compared, as shown in Table 1. No obvious difference was found in the mean age, sex ratio, smoking history, diabetes history, and hypertension history between EC patients and normal controls ().

3.2. Detection of DNA Methylase Levels in EC Patients

The differences in the expression levels of DNA methylases DNMT1, DNMT3a, DNMT3b, and HDAC1 in the EC and healthy people were detected, as shown in Figure 2. Levels of DNMT1, DNMT3a, DNMT3b, and HDAC1 in healthy people were μg/L, μg/L, μg/L, and μg/L, respectively, while those of EC patients were μg/L, μg/L, μg/L, and μg/L, respectively. The expression levels of DNMT1, DNMT3a, DNMT3b, and HDAC of EC patients were much higher ().

(a)

(b)

(c)

(d)

3.3. M6A Gene Methylation Analysis of EC Patients

The differences in methylation rates of ALKBH5, HNRNPC, METTL3, METTL14, WTAP, RBM15, YTHDC1, YTHDF1, YTHDF2, and FTO in EC cancer tissues and adjacent normal tissues were compared, as shown in Figure 3. It was found that the methylation rates of ALKBH5, HNRNPC, METTL3, METTL14, WTAP, RBM15, YTHDC1, YTHDF1, YTHDF2, and FTO in adjacent normal tissues were sharply lower (). The methylation rates of ALKBH5, HNRNPC, METTL3, WTAP, RBM15, YTHDC1, YTHDF1, and FTO in EC tissues were observably higher (). However, no great difference was observed between the methylation rate and unmethylation rate of METTL14 and YTHDF2 ().

The differences in ALKBH5, HNRNPC, METTL3, METTL14, WTAP, RBM15, YTHDC1, YTHDF1, YTHDF2, and FTO methylation levels in tumor tissues and adjacent normal tissues of EC patients were compared, as shown in Figure 4. The ALKBH5, HNRNPC, METTL3, WTAP, RBM15, YTHDC1, YTHDF1, and FTO methylation levels in EC tissues were higher (). However, no visible difference was observed in METTL14 and YTHDF2 methylation levels between EC tissues and adjacent normal tissues ().

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

3.4. IRF Analysis of EC Prognostic Impact

Univariate COX regression analysis of risk factors affecting the survival rate of EC patients is shown in Table 2. There were notable differences in gender, tumor grade, TNM stage, degree of infiltration, and methylation of ALKBH5, HNRNPC, METTL3, and YTHDC1 genes among EC patients with different survival rates (). The survival rate of EC patients with different ages and METTL14, WTAP, RBM15, YTHDF1, YTHDF2, and FTO methylation levels showed no obvious difference ().

According to the median risk assessment in TCGA database, EC patients were rolled into high-risk and low-risk groups, and Kaplan-Meier curves were used to analyze the differences in survival between EC patients with different risks, as shown in Figure 5. The overall survival of patients in the high-risk group of EC was shorter in contrast to that in the low-risk group (). The AUC of the M6A gene methylation prediction model was 0.803, suggesting that the model had good sensitivity and specificity.

(a)

(b)

Multivariate COX regression analysis of IRF affecting the prognosis of EC patients is shown in Table 3. The results showed that gender, tumor grade, TNM grade, degree of infiltration, ALKBH5, HNRNPC, and METTL3 were the IRFs that affected the prognosis of EC patients (), while YTHDC1 was not ().

3.5. Validation of the Improved Fuzzy Data Mining Algorithm

Firstly, was adopted to verify the computational performance of the PSO algorithm used in this work. The maximum number of iterations in the PSO algorithm was set to 1000, the initial population size was 30, and the optimal solution of the algorithm was analyzed in 10, 30, 50, and 100 dimensions, as shown in Figure 6. It was found that with the increase of the algorithm dimension, the error of the PSO algorithm in searching for the optimal solution gradually increased, but the overall solution accuracy of the algorithm model was better. The PSO algorithm was employed to optimize the FCM algorithm. It can be found that when the optimized IFCM algorithm was used for testing on the Glass data set, the error was , while it was before the optimization. It showed that using the PSO algorithm to optimize the FCM algorithm can obtain more accurate prediction results.

(a)

(b)

3.6. EC Prognosis Prediction Evaluation Using the Data Mining Model

The differences between the effects of different data mining models for the prognosis prediction and evaluation of EC patients were compared, as shown in Figure 7. The diagnostic Se, Sp, Acc, PPV, NPV, and AUC of LRA, DT C5.0, ANN, SVM, and IFCM algorithms were 68.8%, 83.8%, 76.3%, 80.9%, 72.8%, and 0.813; 63.8%, 91.3%, 77.5%, 87.9%, 71.6%, and 0.857; 86.3%, 65.0%, 75.6%, 71.1%, 82.5%, and 0.895; 88.8%, 85.0%, 86.9%, 85.5%, 88.3%, and 0.926; and 93.8%, 90.0%, 91.9%, 90.4%, 93.5%, and 0.958, respectively. It can be observed that the diagnostic effect of the IFCM model was the best.

(a)

(b)

4. Discussion

EC is a malignant tumor of the digestive tract due to esophageal epithelial cell lesions. Malnutrition, high eating temperature, smoking, and drinking are the main risk factors for EC, in addition to obesity, gastroesophageal reflux, advanced age, etc. [17]. A large amount of clinical evidence shows that the overall survival rate of EC patients is low, and the 5-year survival rate of patients is only 10% to 19% [18, 19]. In the process of EC, epigenetic changes can lead to the activation of protooncogenes and the inactivation of tumor suppressor genes. DNA methylation is the main form of epigenetic modification, which is mainly the methylation of cytosine in the CpG sequence of gene promoter region to regulate gene expression level [20]. Therefore, DNA methylation is the main mechanism leading to the silencing/inactivation of tumor suppressor genes. The methylation process of DNA requires the catalysis of DNA methyltransferases (DNMT1, DNMT2, DNMT3a, and DNMT3b), transfers the methyl group from s-adenosylmethionine to the five carbon atoms of the cytosine ring, and finally forms DNA methylation [21]. HDACs also play an important role in tumor progression. When the expression of HDACs is upregulated, gene expression will be downregulated due to changes in chromatin structure, thereby affecting tumor progression [22]. In this work, the differences between the protein levels of DNMT1, DNMT3a, DNMT3b, and HDAC1 in peripheral blood of EC patients and healthy people were detected, and it was found that the protein levels of DNMT1, DNMT3a, DNMT3b, and HDAC1 in peripheral blood of EC patients were significantly increased. It indicated that the high methylation level of related genes in the peripheral blood of EC patients may be involved in the disease process.

M6A is a very common methylation modification in eukaryotic mRNA. More and more studies have confirmed that M6A can regulate RNA stability, localization, splicing, and translation at the posttranscriptional level [23, 24]. M6A methylation is a dynamic and reversible process, which is regulated by three main types of methylation regulators, namely, methyltransferases, demethylases, and methylation-binding proteins. Methyltransferase can promote the methylation modification of RNA by M6A, and its encoding genes include METLL3, METLL14, WTAP, and RBM15 [25]. Demethylase can remove the M6A group in RNA and maintain the M6A methylation in a dynamic and reversible transition, and its main encoding genes include ALKBH5 and FTO [26]. Methylation-binding proteins can bind to and function at M6A methylation sites in RNA, and their main encoding genes include YTHDC1, YTHDC2, YTHDF1, and HNRNPC [27]. M6A methylation modification plays an important role in the regulation of gene expression, and the abnormality of its regulatory mechanism may be involved in the occurrence and development of disease or cancer. This work searched TCGA database to download and extract gene expression data for 13 EC-related M6A methylation regulators. Bioinformatics technology and analysis of test results showed that ALKBH5, HNRNPC, METTL3, WTAP, RBM15, YTHDC1, YTHDF1, and FTO methylation levels in EC patient tissues were significantly higher than those in adjacent normal tissues. It indicated that the level of M6A methylation regulators was closely related to the occurrence and development of EC. Then, univariate and multivariate COX regression analysis was performed to affect the prognosis of EC patients, and it was found that gender, tumor grade, TNM grade, degree of infiltration, ALKBH5, HNRNPC, and METTL3 were the IRFs that affected the prognosis of EC patients. Tumor degree of infiltration is closely related to lymph node metastasis, which can be used in EC lymph node metastasis and prognosis evaluation [28]. The perioperative immune function of female EC patients is often better than that of males, and the postoperative quality of life of female EC patients is better, so gender is a protective factor for the prognosis of EC [29]. EC patients with high ALKBH5 expression have a better prognosis because ALKBH5 is able to inhibit the initiating ability of malignancy. METTL3 can promote the growth, survival, and invasion of lung adenocarcinoma cells. METTL3 deletion can change the enrichment state of M6A and promote the growth, self-renewal, and tumor formation of glioma cells [30]. HNRNPC expression is upregulated in cancer cells, which can control endogenous dsRNA and lower-level interferon response in breast cancer cells and then participate in the disease process [31]. In this work, a prognostic model of EC patients was constructed and it was found that the level of M6A methylation regulators was closely related to the progression of the disease, so it can be undertaken as a molecular marker for the early diagnosis and prognosis prediction of EC for further research.

The combined detection of molecular markers can be used for early auxiliary diagnosis of EC, as well as for follow-up and recurrence detection after treatment. To find the deep relationship between variables, data mining technology can solve such multiparameter problems. Medical data has the characteristics of diversity, complexity, redundancy, irregularity, and timeliness. It is impossible to find the hidden laws of data only by relying on the clinical experience of doctors and traditional data statistics methods [32]. Data mining technology provides methodological support, which can combine computers, ANN, and cancer diagnosis to build an intelligent expert system, so it has become a research hotspot in the medical field. This work used logistic regression model, DT C5.0 model, ANN model, SVM model, and IFCM model to predict the prognosis of EC. The results showed that the predicted Se of the above model was 68.8%, 63.8%, 86.3%, 88.8%, and 93.5%, respectively; the Sp was 83.8%, 91.3%, 65.0%, 85.0%, and 90.0%, respectively; the Acc was 76.3%, 77.5%, 75.6%, 86.9%, and 91.9%, respectively; and the AUC was 0.813, 0.857, 0.895, 0.926, and 0.958, respectively. The logistic regression model needs to determine the effect of each variable based on the OR value of the independent variables in the multivariate analysis, so its prediction efficiency is relatively low [33]. ANN model has strong learning ability and fault tolerance ability, and its ability to identify and predict new samples is outstanding, and it is better than the logistic regression model [34]. Using the PSO algorithm to improve FCM can improve the global search ability of the algorithm, reduce the scale difference between data objects, avoid the FCM algorithm falling into local extreme values, and reduce the sensitivity of the algorithm to the initial value setting [35]. In this work, the IFCM algorithm was employed to predict the prognosis of EC, which had stronger model division ability and better clustering effect.

5. Conclusions

The levels of DNA methyltransferase in peripheral blood of EC patients were significantly increased, and the levels of M6A methylation regulators ALKBH5, HNRNPC, METTL3, WTAP, RBM15, YTHDC1, YTHDF1, and FTO were increased in EC tissues. Tumor grade, TNM grade, degree of infiltration, and M6A methylation regulators ALKBH5, HNRNPC, and METTL3 were IRFs for EC prognosis, and gender was a protective factor. The integration of M6A methylation regulators and clinical gender and other information combined with data mining technology for EC early diagnosis and prognosis prediction model showed good accuracy. This work only used the M6A methylation regulator data in TCGA database to construct and validate the EC patient prognosis evaluation model. Future research needed to further explore the CpG islands in the promoter regions of each factor to further improve the results of this work. This work can provide reference materials for the early diagnosis and prognosis evaluation of EC.

Data Availability

The data used to support the findings of this study are included within the article.

Conflicts of Interest

The authors declare that they have no conflicts of interest.

References

F. L. Huang and S. J. Yu, “Esophageal cancer: risk factors, genetic association, and treatment,” Asian Journal of Surgery, vol. 41, no. 3, pp. 210–215, 2018.
View at: Publisher Site | Google Scholar
D. J. Uhlenhopp, E. O. Then, T. Sunkara, and V. Gaduputi, “Epidemiology of esophageal cancer: update in global trends, etiology and risk factors,” Clinical Journal of Gastroenterology, vol. 13, no. 6, pp. 1010–1021, 2020.
View at: Publisher Site | Google Scholar
T. Jordan, D. M. Mastnak, N. Palamar, and N. R. Kozjek, “Nutritional therapy for patients with esophageal cancer,” Nutrition and Cancer, vol. 70, no. 1, pp. 23–29, 2018.
View at: Publisher Site | Google Scholar
L. Zhang, Q. Lu, and C. Chang, “Epigenetics in health and disease,” Advances in Experimental Medicine and Biology, vol. 1253, pp. 3–55, 2020.
View at: Publisher Site | Google Scholar
P. Kaliman, “Epigenetics and meditation,” Current Opinion in Psychology, vol. 28, pp. 76–80, 2019.
View at: Publisher Site | Google Scholar
L. Villanueva, D. Álvarez-Errico, and M. Esteller, “The contribution of epigenetics to cancer immunotherapy,” Trends in Immunology, vol. 41, no. 8, pp. 676–691, 2020.
View at: Publisher Site | Google Scholar
T. Sun, R. Wu, and L. Ming, “The role of m6A RNA methylation in cancer,” Biomedecine & Pharmacotherapie, vol. 112, article 108613, 2019.
View at: Publisher Site | Google Scholar
S. Ma, C. Chen, X. Ji et al., “The interplay between m6A RNA methylation and noncoding RNA in cancer,” Journal of Hematology & Oncology, vol. 12, no. 1, p. 121, 2019.
View at: Publisher Site | Google Scholar
X. Guo, K. Li, W. Jiang et al., “RNA demethylase ALKBH5 prevents pancreatic cancer progression by posttranscriptional activation of PER1 in an m6A-YTHDF2-dependent manner,” Molecular Cancer, vol. 19, no. 1, p. 91, 2020.
View at: Publisher Site | Google Scholar
J. Han, J. Z. Wang, X. Yang et al., “METTL3 promote tumor proliferation of bladder cancer by accelerating pri-miR221/222 maturation in m6A-dependent manner,” Molecular Cancer, vol. 18, no. 1, p. 110, 2019.
View at: Publisher Site | Google Scholar
T. Liu, Q. Wei, J. Jin et al., “The m6A reader YTHDF1 promotes ovarian cancer progression via augmenting EIF3C translation,” Nucleic Acids Research, vol. 48, no. 7, pp. 3816–3831, 2020.
View at: Publisher Site | Google Scholar
S. Müller, M. Glaß, A. K. Singh et al., “IGF2BP1 promotes SRF-dependent transcription in cancer in a m6A- and miRNA-dependent manner,” Nucleic Acids Research, vol. 47, no. 1, pp. 375–390, 2019.
View at: Publisher Site | Google Scholar
K. Makino, S. Lee, S. Bae et al., “Simplified decision-tree algorithm to predict falls for community-dwelling older adults,” Journal of Clinical Medicine, vol. 10, no. 21, p. 5184, 2021.
View at: Publisher Site | Google Scholar
X. Zhou, Y. Li, and W. Liang, “CNN-RNN based intelligent recommendation for online medical pre-diagnosis support,” IEEE/ACM Transactions on Computational Biology and Bioinformatics, vol. 18, no. 3, pp. 912–921, 2021.
View at: Publisher Site | Google Scholar
Z. Wan, Y. Dong, Z. Yu, H. Lv, and Z. Lv, “Semi-supervised support vector machine for digital twins based brain image fusion,” Frontiers in Neuroscience, vol. 15, article 705323, 2021.
View at: Publisher Site | Google Scholar
J. Yin, H. Chang, D. Wang, H. Li, and A. Yin, “Fuzzy C-means clustering algorithm-based magnetic resonance imaging image segmentation for analyzing the effect of edaravone on the vascular endothelial function in patients with acute cerebral infarction,” Contrast Media & Molecular Imaging, vol. 2021, article 4080305, 2021.
View at: Publisher Site | Google Scholar
A. Ashok, D. Niyogi, P. Ranganathan et al., “The enhanced recovery after surgery (ERAS) protocol to promote recovery following esophageal cancer resection,” Surgery Today, vol. 50, no. 4, pp. 323–334, 2020.
View at: Publisher Site | Google Scholar
H. He, N. Chen, Y. Hou et al., “Trends in the incidence and survival of patients with esophageal cancer: a SEER database analysis,” Thoracic Cancer, vol. 11, no. 5, pp. 1121–1128, 2020.
View at: Publisher Site | Google Scholar
G. I. Moral Moral, M. Viana Miguel, Ó. Vidal Doce et al., “Complicaciones postoperatorias y supervivencia del cancer de esofago: analisis de dos periodos distintos,” Cirugía Española (English Edition), vol. 96, no. 8, pp. 473–481, 2018.
View at: Publisher Site | Google Scholar
S. Horvath and K. Raj, “DNA methylation-based biomarkers and the epigenetic clock theory of ageing,” Nature Reviews Genetics, vol. 19, no. 6, pp. 371–384, 2018.
View at: Publisher Site | Google Scholar
H. Takeshima, T. Niwa, S. Yamashita et al., “TET repression and increased DNMT activity synergistically induce aberrant DNA methylation,” The Journal of Clinical Investigation, vol. 130, no. 10, pp. 5370–5379, 2020.
View at: Publisher Site | Google Scholar
M. Edderkaoui, C. Chheda, B. Soufi et al., “An inhibitor of GSK3B and HDACs kills pancreatic cancer cells and slows pancreatic tumor growth and metastasis in mice,” Gastroenterology, vol. 155, no. 6, pp. 1985–1998.e5, 2018.
View at: Publisher Site | Google Scholar
H. Zhang, X. Shi, T. Huang et al., “Dynamic landscape and evolution of m6A methylation in human,” Nucleic Acids Research, vol. 48, no. 11, pp. 6251–6264, 2020.
View at: Publisher Site | Google Scholar
S. Oerum, V. Meynier, M. Catala, and C. Tisné, “A comprehensive review of m6A/m6Am RNA methyltransferase structures,” Nucleic Acids Research, vol. 49, no. 13, pp. 7239–7255, 2021.
View at: Publisher Site | Google Scholar
S. Oerum, M. Catala, C. Atdjian et al., “Bisubstrate analogues as structural tools to investigate m6A methyltransferase active sites,” RNA Biology, vol. 16, no. 6, pp. 798–808, 2019.
View at: Publisher Site | Google Scholar
S. Nie, L. Zhang, J. Liu et al., “ALKBH5-HOXA10 loop-mediated JAK2 m6A demethylation and cisplatin resistance in epithelial ovarian cancer,” Journal of Experimental & Clinical Cancer Research : CR, vol. 40, no. 1, p. 284, 2021.
View at: Publisher Site | Google Scholar
S. D. Kasowitz, J. Ma, S. J. Anderson et al., “Nuclear m6A reader YTHDC1 regulates alternative polyadenylation and splicing during mouse oocyte development,” PLOS Genetics, vol. 14, no. 5, article e1007412, 2018.
View at: Publisher Site | Google Scholar
Q. Y. Chen, Y. N. Li, X. Y. Wang et al., “Tumor fibroblast-derived FGF2 regulates expression of SPRY1 in esophageal tumor-infiltrating T cells and plays a role in T-cell exhaustion,” Cancer Research, vol. 80, no. 24, pp. 5583–5596, 2020.
View at: Publisher Site | Google Scholar
S. Li, H. Chen, J. Man et al., “Changing trends in the disease burden of esophageal cancer in China from 1990 to 2017 and its predicted level in 25 years,” Cancer Medicine, vol. 10, no. 5, pp. 1889–1899, 2021.
View at: Publisher Site | Google Scholar
A. Visvanathan, V. Patil, A. Arora et al., “Essential role of METTL3-mediated m⁶A modification in glioma stem-like cells maintenance and radioresistance,” Oncogene, vol. 37, no. 4, pp. 522–533, 2018.
View at: Publisher Site | Google Scholar
Y. Wu, W. Zhao, Y. Liu et al., “Function of HNRNPC in breast cancer cells by controlling the dsRNA-induced interferon response,” The EMBO Journal, vol. 37, no. 23, article e99017, 2018.
View at: Publisher Site | Google Scholar
S. Xie, Z. Yu, and Z. Lv, “Multi-disease prediction based on deep learning: a survey,” CMES-Computer Modeling in Engineering and Sciences, vol. 128, no. 2, pp. 489–522, 2021.
View at: Publisher Site | Google Scholar
W. Sun, Z. Cai, Y. Li, F. Liu, S. Fang, and G. Wang, “Data processing and text mining technologies on electronic medical records: a review,” Journal of Healthcare Engineering, vol. 2018, Article ID 4302425, 9 pages, 2018.
View at: Publisher Site | Google Scholar
A. Arefinia, O. Bozorg-Haddad, A. Oliazadeh, and H. A. Loáiciga, “Reservoir water quality simulation with data mining models,” Environmental Monitoring and Assessment, vol. 192, no. 7, p. 482, 2020.
View at: Publisher Site | Google Scholar
J. B. Raja and S. C. Pandian, “PSO-FCM based data mining model to predict diabetic disease,” Computer Methods and Programs in Biomedicine, vol. 196, article 105659, 2020.
View at: Publisher Site | Google Scholar

Copyright

Copyright © 2022 Shuoming Wu et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies