Detection of Cervical Cancer Using lncRNA Expression Data and Detection of Possible Biomarkers with Bagged CART Machine Learning Method

Abstract

Aim: Cervical cancer (CC), one of the most common gynecological cancers, occurs when the cell layer that forms the surface of the cervix turns into abnormal cells. This type of cancer ranks fourth in cancer-related female deaths, and 3.6% of women living in developed countries suffer from this disease, while approximately 15% of women living in underdeveloped countries are exposed to this cancer. The primary means of reducing the high mortality associated with the disease is early diagnosis and treatment. Especially in underdeveloped countries and in countries where screening programs are not adequately established, early diagnosis of cancer and initiation of rapid and effective treatment are very important in reducing possible deaths and increasing survival rates. Therefore, new studies and biomarkers for the early detection of this cancer are needed. Therefore, within the scope of this study, using the lncRNA expression data of patients with open access CC and paracancerous tissues, the data were classified with Bagged CART, which is one of the machine learning (ML) methods, and biomarkers that may be associated with cancer were obtained as a result of modeling. Methods: In the current study, an open-access dataset was used to reveal the relationship of lncRNAs with CC. The dataset includes data from samples taken from 9 paracancerous tissues with 9 cervical cancers. In the modeling phase, the Bagged CART method was applied using 5-fold cross-validation. Performance values obtained as a result of the model were evaluated with accuracy (ACC), balanced accuracy (b-ACC), sensitivity (SE), specificity (SP), positive predictive value (ppv), negative predictive value (npv), and F1-score. Results: In the study, LASSO variable selection method was used in order to select the most important variables associated with the output variable and to reduce the number of input variables. Modeling was done with 14 lncRNAs selected by this method. When the performance metrics obtained by the modeling were examined, ACC, b-ACC, SE, SP, ppv, npv and F1-score were obtained as 94.4%, 94.4%, 100%, 88.9%, 90%, 100%, 94.7%, respectively. When the variable significance values obtained from the bagged CART results were examined, it was seen that the lncRNAs that most explained CC were RP11-80I15.4, RP11-183I6.2, AC144525.1, RP11-129B22.1, and RP11-389K14.3. Conclusion: When the findings obtained from the study were examined, possible biomarker lncRNAs for CC were detected using Bagged CART, one of the ML methods. With comprehensive analyzes on the subject, more accurate and reliable results can be obtained and the reliability of possible marker lncRNAs can be tested.