Integrating temporal single-cell gene expression modalities for trajectory inference and disease prediction

Abstract

Current methods for analyzing single-cell datasets have relied primarily on static gene expression measurements to characterize the molecular state of individual cells. However, capturing temporal changes in cell state is crucial for the interpretation of dynamic phenotypes such as the cell cycle, development, or disease progression. RNA velocity infers the direction and speed of transcriptional changes in individual cells, yet it is unclear how these temporal gene expression modalities may be leveraged for predictive modeling of cellular dynamics. Here, we present the first task-oriented benchmarking study that investigates integration of temporal sequencing modalities for dynamic cell state prediction. We benchmark eight integration approaches on eight datasets spanning different biological contexts, sequencing technologies, and species. We find that integrated data more accurately infers biological trajectories and achieves increased performance on classifying cells according to perturbation and disease states. Furthermore, we show that simple concatenation of spliced and unspliced molecules performs consistently well on classification tasks and can be used over more memory intensive and computationally expensive methods. This work provides users with practical recommendations for task-specific integration of single-cell gene expression modalities.

Inferring single-cell transcriptomic dynamics with structured latent gene expression dynamics

“…We compare our model with existing batch correction tools for gene-expression (without velocity). Recently a method using traditional batch correction tools that produce a corrected gene count matrix has been suggested for RNA velocity [29, 30]. We compare this method to LatentVelo by using UniTVelo unified time mode combined with batch correction tools for gene counts.…”

“…Our model is the only method scoring highly on all metrics, even when compared to these methods specifically designed for batch correction of gene expression. In d), we evaluate batch correction of velocity, and compare with UniTVelo on the unintegrated data or using a previously developed method for RNA velocity (see methods) with batch correction methods that output corrected gene-expression matrices (ComBat [31], scGen [32]) [29, 30]. These methods fail to properly integrate the batches and have worse performance on velocity metrics than our model.…”

“…In d), we evaluate batch correction of velocity. We compare LatentVelo with UniTVelo used on the unintegrated data, or UniTVelo used with a previously suggested method for batch correction of RNA velocity [32, 33] (see Methods section V H for details). This approach requires the use of any generic batch correction method that output corrected gene-expression matrices, and so we use this method with ComBat [34] (as was originally suggested), and scGen [35] (an auto encoder method).…”

Integrating temporal single-cell gene expression modalities for trajectory inference and disease prediction

“…github. com/ jranek/ EVI [96] and in the Zenodo repository [97]. Source code is released under the MIT license.…”

“…Loom files and preprocessed data are available in the Zenodo repository https://doi.org/10.5281/zenodo.6587903 [ 95 ]. Source code including all functions for preprocessing, integration, and evaluation are publicly available at www.github.com/jranek/EVI [ 96 ] and in the Zenodo repository [ 97 ]. Source code is released under the MIT license.…”