Evaluating Reproductive Potential Of Human Embryos Using Molecular Spectroscopy Combined With Chemometric Methods

Thousands of families are benefit from the development of in vitro fertilization and embryo transfer (IVF-ET) technology. However, only30%of IVF cycles result in a pregnancy. Therefore, multiple embryos are replaced per treatment cycle to increase pregnancy rates. This has sharply increased multiple pregnancy rate and put patients under pressure both mentally and economically.A key step in IVF is the assessment of embryo viability in order to identify the embryos most likely to result in pregnancy. Currently used embryo assessment systems are largely based on morphology. While these systems have been pivotal in improving implantation rates and reducing multiple pregnancy rates, their precision is still insufficient. It is still necessary to find a rapid and effective way to evaluate the reproductive potential of embryos, in order to solve the two biggest problems mentioned above.In this paper, molecular spectroscopy, such as raman, fluorescence, ultraviolet-visible and near infrared (NIR), were used to determine the concentration of metabolomics in embryo culture media. And different chemometric methods were used to establish a discriminative model to evaluate the reproductive potential of human embryos. Results showed that model established using NIR spectra combined with partial least square-discriminant analysis (PLS-DA) method owns the highest correct classification rate. This model can distinguish embryos that lead to delivery from those can not. And after using mean centering and Savitzky-Golay2nd derivation algorithm, the correct classification rate of prediction set in PLS-DA model raised to77%. The sensitivity and specificity of prediction set were88%and70%respectively.In order to improve performance of the model, three variable selection methods, including Monte Carlo uninformative variable elimination (MC-UVE), competitive adaptive reweighted sampling (CARS), stability competitive adaptive reweighted sampling (SCARS), were applied concerning establishment of a robust PLS-DA model. The results showed that, correct classification rates were increased to74.24%,77.12%and80.10%respectively after using the three methods, comparing with74%of using full spectra. And number of the selected variables reduced to less than50. Compared with other two algorithms, SCARS displayed the best result in model optimization and stabilization. The experiments mentioned above all confirmed the fact that molecular spectroscopy exhibited a potential application in evaluation of the reproductive potential of human embryos.