Multi-Omics Analysis and Machine Learning Prediction Model for Pregnancy Outcomes After Intracytoplasmic Sperm Injection–in vitro Fertilization

Abstract

Background: To explore the methylation profiles in cumulus cells (CCs) of women undergoing intracytoplasmic sperm injection–in vitro fertilization (ICSI–IVF) and establish a prediction model of pregnancy outcomes using machine learning approaches. Methods: Methylation data were retrieved from the Gene Expression Omnibus (GEO) database, and differentially methylated genes (DMGs) were subjected to gene set analyses. Support vector machine (SVM), random forest (RF), and logistic regression (LR) were used to establish the prediction model, and microarray data from GEO was analyzed to identify differentially expressed genes (DEGs) associated with the dichotomous outcomes of clinical pregnancy (pregnant vs. non-pregnant). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis provided multi-dimensional validation for selected DMGs. Results: A total of 338 differentially methylated CpG sites associated with 146 unique genes across the genome were identified. Among the identified pathways, the prominent ones were involved in the regulation of cell growth and oocyte development (hsa04340, hsa04012, hsa04914, hsa04614, hsa04913, hsa04020, and hsa00510). The area under the curve (AUC) of machine learning classifiers was 0.94 (SVM) vs. 0.88 (RF) vs. 0.97 (LR). 196 DEGs were found in transcriptional microarray. Mapped genes were selected through overlapping enriched pathways in transcriptional profiles and methylated data of CCs, predictive of successful pregnancy. Conclusions: Methylated profiles of CCs were significantly different between women receiving ICSI-IVF procedures that conceived successfully and those that did not conceive. Machine learning approaches are powerful tools that may provide crucial information for prognostic assessment. Pathway analysis may be another way in multiomics analysis of cumulus cells.