P–275 Development of a prediction model using machine learning on small noncoding RNA biomarkers for non-invasive selection of high-quality embryos for the in vitro fertilization process

Study question The aim of the study was to identify molecules in the embryo culture medium as important predictive biomarkers of high-quality embryos Summary answer The study identified 14 canonical iso-miRNA molecules that prognostically determine the quality of the embryo with a prediction accuracy with 95% sensitivity and 80% specificity. What is known already The quality of the embryo for the success of the IVF process is not specifically diagnosed, only morphological features (monitoring in the embryoscope) are considered. Embryo quality selection systems have likely reached their peak. The success rate of the IVF process is only 29%.; it is therefore necessary to look for other biomarkers. The oocyte itself can significantly predict the development of the early embryo,as it is a supplier of RNA and cellular mechanisms. However, collection follicular fluid is technically demanding. The probability of oocyte fertilization does not reach the required percentage therefore other embryological techniques multiply the economic costs. Study design, size, duration Women (n = 734) who visited the IVF centre were recruited for the study. Oocytes were collected from 54 of them and used for IVF. After 4/5-day embryo cultivation, the best quality embryo was selected and used for implantation into the uterus. The culture medium has been collected from 60 embryos during 3 years (2018–2020). Written informed consent was obtained from all patients. The study has been approved by the Ethical committee of the Košice governing region Participants/materials, setting, methods We used fresh/frozen culture media of embryos selected using an embryoscope. Further, information regarding the success of IVF, pregnancy and IVF failure was collected. Culture media libraries of noncoding small RNAs (miRNAs) were examined using massively parallel sequencing on the Illumina platform. Obtained data was processed with freely available bioinformatic tools and machine learning. For methods with different models, the number of predictive biomarkers and specific prognostic-predictive molecules were selected. Main results and the role of chance The main results of the study specifically identify ncRNA molecules that prognostically and predictively select a high-quality embryo suitable for IVF transmission from a low-quality embryo with 95% sensitivity and 80% specificity with an average accuracy of 85% in 4 different models. We also determined the minimum of 14 miRNA as prediction biomarkers. The developed model can predict embryo quality from the culture medium based on ncRNA results from sequence data and set the cut-off value for the expression and significance of individual miRNA molecules with respect to embryo quality. Furthermore, positive and negative correlations of miRNA molecules with different distributions in a high-quality embryo compared to a low-quality embryo were determined. The molecules identified in the embryo culture medium were organized according to their importance, resp. significance based on their significance coefficient. So far, there is no evidence of pending patents regarding the distribution of specific canonical miRNAs and iso-miRNA molecules analysed by massively parallel sequencing in terms of biological competence and embryo quality determination with multifactorial consideration of its variation. This is the first study focused on the success of the IVF process based on embryo quality prediction. Limitations, reasons for caution Exploratory data need to be validated in a larger scale study. Wider implications of the findings: The given miRNA molecules and the software model can be used as a safe,non-invasive diagnostic test for the selection of a highly competent embryo. Canonical and iso-miRNA molecules from the study can be used in other forms of diagnostic assays, such as specific embryo selection probes and, plate hybridization assay. Trial registration number non clinical trials