Linking single-neuron morphology and gene expression using deep learning

点击翻译按钮获取中文摘要

Abstract Characterizing neuronal cell identity in terms of transcriptomics, electrophysiology, and morphology is an essential component for understanding neural circuits and function. A multi-modal understanding of cell-to-cell variation in both single-cell transcriptional profiles and morphological phenotypes is needed to understand functional characteristics and the emergence of complexities in the brain. High-throughput single-cell measurements of neuronal gene expression are available, but the relationship between morphology and gene expression is not well explored due to the challenge of measuring both modalities from the same cells. We hypothesize that gene expression influences neuronal morphology, and thus that single-cell gene expression can be used to predict single-cell morphology. We propose to leverage recent advances in deep generative models to predict the distribution of single-cell morphology images from single-cell gene expression. We will: (1) develop deep generative to learn the relationship between single-cell gene expression and morphology; (2) train MorphGAN to generate morphologies for unseen neuronal single-cell gene expression profiles.; and (3) identify morphological axes of variation and key genes that predict morphology using MorphGAN. Completion of these aims will produce publicly available software tools and a public database of predicted single-cell neuron morphology images. Ultimately, linking transcriptomic and morphological characteristics of single neurons would be invaluable in capturing the diversity of brain cells and delineating neuronal cell types.