变分因子分析:一种基于深度生成模型的单细胞转录组数据因子分析方法

VFA:a factor analysis algorithm for single-cell transcriptomic data based on the deep generative model

目的提出一种基于深度生成模型的单细胞转录组数据因子分析方法,并通过从细胞和基因两个层面探索生物异质性来解释模型结果。方法采用深度生成网络,构建面向单细胞转录组数据因子分析的深度生成模型,用小鼠胚胎发育细胞转录组数据训练模型,对结果进行细胞类型注释、因子及载荷可视化等下游生物学分析。结果该模型实现了单细胞转录组数据的降维并同时保留了异质性,通过因子识别了不同细胞类型之间的特征,通过载荷鉴定出细胞特异性基因。结论基于深度生成模型构建的因子分析方法具有更好的可解释性,能从信息复杂、具有错综关系的单细胞转录组数据中提取代表性信息,并从不同层面解析其生物学含义,为单细胞数据分析开拓了新思路。

Objective To propose a deep generative model based factor analysis method for single-cell transcriptomic data,and explain the model results by exploring biological heterogeneity from both cell-and gene-levels.Methods A deep generative model for factor analysis of single-cell transcriptomics data was established,which was trained using the transcriptomic data of mouse embryonic developmental cells.Downstream biological analysis on the results was performed,such as annotation of cell types and visualization of factors and loadings.Results The model achieved dimensionality reduction of single-cell transcriptomic data while retaining the heterogeneity.The factors obtained from model training recognized features of different cell types,and the factor loadings identified cell-type-specific genes.Conclusion This factor analysis method based on the deep generative model has better interpretability,which can extract representative information from complex and intricate single-cell transcriptomic data,and interpret biological meanings at different levels.