Zheng Wei, Zhang Shuoping, Gu Yifan, Gong Fei, Kong Lingyin, Lu Guangxiu, Lin Ge, Liang Bo, Hu Liang
National Health Commission (NHC) Key Laboratory of Human Stem Cell and Reproductive Engineering, School of Basic Medical Science, Institute of Reproductive and Stem Cell Engineering, Central South University, Changsha, China.
Clinical Research Center for Reproduction and Genetics in Hunan Province, Reproductive and Genetic Hospital of CITIC-Xangya, Changsha, China.
Front Physiol. 2021 Nov 19;12:777259. doi: 10.3389/fphys.2021.777259. eCollection 2021.
This study aimed to establish a non-invasive predicting model via Raman spectroscopy for evaluating the blastocyst development potential of day 3 high-quality cleavage stage embryos. Raman spectroscopy was used to detect the metabolic spectrum of spent day 3 (D3) embryo culture medium, and a classification model based on deep learning was established to differentiate between embryos that could develop into blastocysts (blastula) and that could not (non-blastula). The full-spectrum data for 80 blastula and 48 non-blastula samples with known blastocyst development potential from 34 patients were collected for this study. The accuracy of the predicting method was 73.53% and the main different Raman shifts between blastula and non-blastula groups were 863.5, 959.5, 1,008, 1,104, 1,200, 1,360, 1,408, and 1,632 cm from 80 blastula and 48 non-blastula samples by the linear discriminant method. This study demonstrated that the developing potential of D3 cleavage stage embryos to the blastocyst stage could be predicted with spent D3 embryo culture medium using Raman spectroscopy with deep learning classification models, and the overall accuracy reached at 73.53%. In the Raman spectroscopy, ribose vibration specific to RNA were found, indicating that the difference between the blastula and non-blastula samples could be due to materials that have similar structure with RNA. This result could be used as a guide for biomarker development of embryo quality assessment in the future.
本研究旨在通过拉曼光谱建立一种非侵入性预测模型，以评估第3天高质量卵裂期胚胎的囊胚发育潜力。使用拉曼光谱检测第3天（D3）废弃胚胎培养基的代谢谱，并建立基于深度学习的分类模型，以区分能够发育成囊胚的胚胎和不能发育成囊胚的胚胎。本研究收集了来自34例患者的80个囊胚样本和48个非囊胚样本的全光谱数据，这些样本具有已知的囊胚发育潜力。通过线性判别法，预测方法的准确率为73.53%，囊胚组和非囊胚组之间主要的不同拉曼位移为863.5、959.5、1008、1104、1200、1360、1408和1632 cm（来自80个囊胚和48个非囊胚样本）。本研究表明，利用拉曼光谱结合深度学习分类模型，通过第3天废弃胚胎培养基可以预测第3天卵裂期胚胎发育到囊胚期的潜力，总体准确率达到73.53%。在拉曼光谱中，发现了RNA特有的核糖振动，这表明囊胚和非囊胚样本之间的差异可能归因于与RNA结构相似的物质。这一结果可为未来胚胎质量评估生物标志物的开发提供指导。
