Abstract: BACKGROUND: The etiology of recurrent spontaneous abortion is complex. With the development of genetics and other fields, it has been found that the abnormal expression of autophagy-related genes may lead to the imbalance of cell homeostasis, thus triggers pathological processes, such as apoptosis, inflammatory response and immunosuppressive response, and affects the endometrial microenvironment, trophoblastic function and immune cell function, thereby leading to recurrent spontaneous abortion. By using the recurrent spontaneous abortion samples in the Gene Expression Omnibus database, the expression changes and regulatory mechanisms of autophagy-related genes were analyzed, which is helpful to reveal the mechanism of recurrent spontaneous abortion and develop new therapeutic strategies. However, the specific mechanism of autophagy-related genes in recurrent spontaneous abortion and their interaction with other biological processes still need to be further studied.
OBJECTIVE: To establish a risk score prognostic model for patients with recurrent spontaneous abortion based on autophagy-related genes.
METHODS: Endometrial gene expression matrix of patients with recurrent abortion was obtained from the Gene Expression Omnibus database, autophagy-related genes were obtained from the Human Autophagy Database (HADb), and 30 differentially co-expressed autophagy-related genes were identified. The biological functions of autophagy-related genes were analyzed by gene ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) and DisGeNET enrichment. LASSO and logistic regression analyses were used to identify 16 autophagy-related genes as potential biomarkers. A Nomogram model was subsequently established for the target gene, and the prediction accuracy of the model was evaluated by using the receiver operating characteristic curve. The optimal machine model was selected by comparing the performance of six machine learning models, including random forest model, support vector machine model and generalized linear model. Nomogram, calibration curve and decision curve were used to analyze the machine model to verify the validity of the prediction.
RESULTS AND CONCLUSION: (1) GO analysis showed that the functions of autophagy-related genes in patients with recurrent abortion were mainly involved in autophagy regulation, cellular catabolism, and the formation of mitochondria or other organelle membranes. (2) KEGG analysis showed that autophagy-related genes were enriched in autophagy regulation, phosphatidylinositol 3-kinase/protein kinase B signaling pathway, human papillomavirus infection and neurodegeneration pathway in patients with recurrent abortion. (3) Gene set enrichment analysis showed that autophagy-related genes in patients with recurrent abortion were involved in the biological processes of copper detoxification and proton transmembrane transport, the cellular components of sarcoplasmic reticulum, and the molecular function of regulating nucleoside diphosphate phosphatase activity. (4) A predictive risk model was constructed, and 16 specific autophagy related genes were found to be predictive targets for recurrent abortion. (5) The best neural network model was selected according to the detection efficiency of the machine model, and five most important gene variables related to recurrent abortion autophagy (MAP2K7, CALCOCO2, SAR1A, TUSC1 and STK11) were identified. (6) Using the European population analysis of recurrent abortion samples in the GEO database, gene expression patterns, differentially expressed genes and signaling pathways in recurrent abortion samples can be analyzed from the level of genetic single nucleotide polymorphisms. The predictive model and machine learning model based on good predictive efficiency can identify new therapeutic targets and potential biomarkers related to autophagy in patients with recurrent abortion, which has certain guiding significance for clinical work and mechanism research.

Key words: recurrent spontaneous abortion, autophagy, single nucleotide polymorphism, prognostic model, machine learning, bioinformatics