Rapid and non-invasive diagnostic techniques for embryonic developmental potential: a metabolomic analysis based on Raman spectroscopy to identify the pregnancy outcomes of IVF-ET.

Abstract

The non-invasive and rapid assessment of the developmental potential of embryos is of great clinical importance in assisted reproductive technology (ART). In this retrospective study, we analyzed the metabolomics of 107 samples provided by volunteers and utilized Raman spectroscopy to detect the substance composition in the discarded culture medium of 53 embryos resulting in successful pregnancies and 54 embryos that did not result in pregnancy after implantation. The culture medium from D3 cleavage-stage embryos was collected after transplantation and a total of 535 (107 × 5) original Raman spectra were obtained. By combining several machine learning methods, we predicted the developmental potential of embryos, and the principal component analysis-convolutional neural network (PCA-CNN) model achieved an accuracy rate of 71.5%. Furthermore, the chemometric algorithm was used to analyze seven amino acid metabolites in the culture medium, and the data showed significant differences in tyrosine, tryptophan, and serine between the pregnancy and non-pregnancy groups. The results suggest that Raman spectroscopy, as a non-invasive and rapid molecular fingerprint detection technology, shows potential for clinical application in assisted reproduction.