Article
Inferring developmental trajectories and optimized dimension reduction from temporal single-cell RNA-sequencing data
Search Medline for
Authors
Published: | September 24, 2021 |
---|
Outline
Text
Single-cell RNA-sequencing data from multiple time points promises insights into mechanisms controlling differentiation and cell fate decisions at the level of individual cells. Yet, typically only a small number of time points are available. Due to the destructive nature of the sequencing protocol, at each time point, a different, heterogeneous sample of cells is obtained, comprising cells from a mixture of cell types and in diverse developmental stages. This complicates the identification of specific developmental trajectories across multiple time points, e.g., when inspecting the corresponding sequence of plots after reducing the dimension of the gene expression data for visualization.
To address this challenge, we propose a modeling approach that integrates neural network-based dimension reduction with inference of the temporal dynamics. More specifically, we use a deep learning approach to infer a low-dimensional, latent representation of gene expression. In this latent space, we optimize a dynamic model to describe trajectories by alternating between assigning cells into groups based on the current dynamic model, and optimizing the model parameters by matching the distributions distribution of the model predictions with the true distribution in each group using a quantile-based loss function.
We couple the dimension reduction step with inference of the dynamics by jointly optimizing the dynamic model and the neural network encoding the data in the low-dimensional space, such that a dimension-reduced representation of gene expression can be found that is specifically suited to model the temporal developments present in the data.
Based on simulated data, we show that this approach allows for inferring distinct developmental trajectories, despite the lack of one-to-one correspondence between individual cells at different time points, while additionally learning an improved low-dimensional representation specifically adapted to the underlying dynamics. We additionally present an application on single-cell RNA-sequencing data from several time points during mouse cortical differentiation, illustrating the potential of the approach to provide insights into the dynamics of developmental processes.
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.