Establishment of an artificial neural network model using immune-infiltration related factors for endometrial receptivity assessment
Abstract Background A comprehensive clinical strategy for infertility involves treatment and, more importantly, post-treatment evaluation. Endometrial receptivity, as a component of assessment, does not have a validated tool. This study was anchored on immune factors, which act as critical factors affecting embryonic implantation, was aimed at establishing novel approaches for assessing endometrial receptivity to guide clinical practice. Methods Immune-infiltration levels in the GSE58144 dataset (n = 115) from GEO were analyzed by digital deconvolution and validated by immunofluorescence (n = 30). Modules that were most associated with M1/M2 macrophages and their hub genes were then selected by weighted gene co-expression network as well as univariate analyses and validated by GSE5099 macrophage dataset and qPCR analysis (n = 16). Finally, the artificial neural network model was established from hub genes and its predictive efficacy was validated in the GSE165004 dataset (n = 72). Results Dysregulation of the ratio of M1 to M2 macrophages is an important factor contributing to decreased endometrial receptivity. Selected hub genes, RPS9, DUT, and KIAA0430, were significantly altered in patients with endometrial receptivity decreased and found relating to M1/M2 through ribosomal and proteasomal pathways, and were significantly altered in patients. The model established by the selected hub genes exhibited an excellent predictive value in both datasets with an accuracy of 98.3% and an AUC of 0.975 (95% CI 0.945-1). Conclusions M1/M2 polarization can influence endometrial receptivity by the three genes regulation, while the established ANN model can effectively assess endometrial receptivity to inform patients' pregnancy and individualized clinical management strategies.
