Application of Autoencoder Interpolation to Enhance the Accuracy of Single-Cell Transcriptomic Analysis in Early Porcine Embryos

Abstract

【Objective】Autoencoder, as one of the deep learning algorithms, offers unique advantages in reducing dimensionality and conducting interpolation analysis on single-cell transcriptome data. The study aims to assess the feasibility of applying the autoencoder AutoClass to early porcine embryo single-cell transcriptome data, and to investigate the impacts of different embryonic activation methods on key genes and signaling pathways.【Method】Single-cell transcriptome data were collected from early porcine embryos with 3 different activation modes (in vivo fertilization, in vitro fertilization, and orphaned females), and AutoClass was used to perform data quality control and interpolation analyses and evaluate AutoClass performance in conjunction with downstream analyses. In addition, an in-depth comparison of key genes and signaling pathways in the three types of embryos was performed by differential expression genes (DEGs) and functional enrichment analysis.【Result】After quality control and autoencoder data interpolation, the accuracy of clustering analysis was enhanced, leading to clear clustering among early embryos with different activation methods. QC alone only screened out 1 287 DEGs, while the number of DEGs increased to 11 523 after autocoder interpolation. Functional enrichment analysis further unearthed key biological processes and signaling pathways that were significantly different among the three different types of embryos, such as basic biological processes and cellular metabolism in porcine in vivo fertilized embryos, gonadal development and sex determination in porcine in vitro fertilized embryos and immune defenses and regulation of secretory pathways in orphaned embryos.【Conclusion】The autoencoder was adopted to improve the analytical accuracy of single-cell transcriptome data, whereby the key genes and signaling pathways affecting early embryonic development in three embryonic activation modes were revealed, which provided new ideas for further in-depth study of the molecular mechanisms underlying early embryonic development in porcine.

Keywords

• porcine
• early embryo
• single-cell transcriptome
• in-dept learning
• autoencoder