Machine Learning Identifies Key Gene Markers Related to Fetal Retina Development at Single-Cell Transcription Level.

Purpose: The retina is part of the central nervous system, and its function is vision. During the embryonic development of the fetus, it diversifies the seven major types of cells from a retinal progenitor cell. The purpose of this study is to investigate essential features that are necessary for the development of the fetal retina. Methods: We generated a comprehensive single-cell transcriptional atlas of the human retina by leveraging datasets from the Chan Zuckerberg Initiative Single-Cell Biology collection of human fetal retinas. The eight critical types of retinal cells were investigated, including amacrine, bipolar, cone, horizontal, Müller glia, retinal ganglion, rod, and retinal progenitor cells. We evaluated a total of 36,503 gene features across three developmental stages (early, middle, and late) for each cell type. Using seven feature ranking algorithms and incremental feature selection method, we identified key gene features, constructed efficient classifiers and classification rules. Results: For amacrine cells, RELN and DAB1 are critical; for bipolar cells, ANK3 and RIMS2; for cone cells, PDE6H and FRMPD2; for horizontal cells, NFIA; for Müller glial cells, WIF1 and TF; for retinal ganglion cells, IL1RAPL2 and PCP4; for rod cells, RIMS2 and NRG1; and for retinal progenitor cells, BTG1. These gene features bring into focus the pattern of gene regulation and developmental pathways of the retinal cells for deeper insights into retinal development. Conclusions: This study explored molecular features related to the development of the fetal retina and their potential roles in certain pathways, which may provide novel insights into retinal development and contribute to a better understanding of other retinal diseases.