Single Cell Velocity Analysis

Single Cell Velocity Analysis estimates velocities for studying cellular dynamics. It takes an Expression Matrix with spliced and unspliced counts (Image expr_matrix_spliced_unspliced_16_n_p) as input and produces a Velocity Matrix (Image velocity_matrix_16_n_p) and Cell Annotations (Image cell_annotations_16_n_p). We recommend normalizing the input with the Normalize Single Cell Data tool (see Normalize Single Cell Data).

The tool can be found in the Toolbox here:

        Gene Expression (Image sc_gene_expression_folder_open_16_n_p) | Velocity Analysis (Image velocity_folder_open_16_n_p) | Single Cell Velocity Analysis (Image sc_velocity_analysis_16_n_p)

The tool offers options to run dimensionality reduction or feature selection prior to velocity estimation. To perform feature selection through highly variable genes (HVGs), the data has to be normalized first with the Normalize Single Cell Data tool. When HVGs are used, velocity is estimated only for these genes. This can greatly speed up the calculations. We recommend using HVGs whenever possible. Note that top velocity genes are not necessarily top HVGs. The default value of $ 2,000$ is a good starting point - a too small value can lead to missing important velocity genes, while a too high value will diminish the computation gain. For details on dimensionality reduction or feature selection, please see Feature selection and dimensionality reduction.

The following additional options are available (figure 8.2):

Image velocity_options
Figure 8.2: The options in the dialog of the Single Cell Velocity Analysis tool.

Multi-sample input: There are no well-established approaches for joint batch correction of spliced and unspliced counts. We recommend caution when analyzing a matrix containing multiple samples. If the matrix is batch corrected using the Normalize Single Cell Data tool (see Normalize Single Cell Data), then the correction is only applied to the total gene expression, which is used for k-nearest neighbor graph construction, and not to the spliced and unspliced counts, which are used for velocity estimation. See [Bergen et al., 2021] for a discussion on batch correction for velocity estimation.

Single nucleus RNA sequencing (snRNA-Seq): Velocity estimation has been developed for scRNA-Seq data and it is yet to be determined how well the method works for snRNA-Seq, where the assumptions of the model might not hold [Bergen et al., 2021]. We recommend caution when analyzing and interpreting the results for this type of data.



Subsections