Kernels compute cell-cell transition probabilities based on various input data modalities, including molecular similarity, RNA velocity [La Manno et al., 2018], experimental time points, and many more. They come with methods for high-level, qualitative visualization, including vector field or random walk plots. For quantitative analysis of kernel-computed transition matrices, we recommend taking look at the estimators.

We don’t use the term “kernel” in the way it is used in mathematics, but rather colloquially, to refer to a class that takes in multi-view single-cell data and outputs a cell-cell transition matrix.

kernels.VelocityKernel(adata[, backward, ...])

Kernel which computes a transition matrix based on RNA velocity.

kernels.ConnectivityKernel(adata[, ...])

Kernel which computes transition probabilities based on similarities among cells.

kernels.PseudotimeKernel(adata, time_key[, ...])

Kernel which computes directed transition probabilities based on a k-NN graph and pseudotime.

kernels.CytoTRACEKernel(adata[, backward])

Kernel which computes directed transition probabilities using the CytoTRACE score [Gulati et al., 2020].

kernels.RealTimeKernel(adata, time_key[, ...])

Kernel which computes transition matrix using optimal transport couplings.

kernels.PrecomputedKernel(object[, adata, ...])

Kernel which is initialized based on a precomputed transition matrix.