Cellrank component

[bio-la]

from velocity_component import velocity_vectors

1. Initialize kernel

from cellrank.tl.kernels import VelocityKernel
vk = VelocityKernel(adata)

input: anndata with velocity vectors
output: initialized kernel object

2. Compute transition matrix

vk.compute_transition_matrix()

input: initialized kernel object
output: kernel object with transition matrix

3. optional but recommended correct velocity kernel by combining with connectivity kernel

ck = ConnectivityKernel(adata).compute_transition_matrix()
#NB 3.1,3.2 are just 1,2 with different param choice

3.1 compute connectivity kernel 
	input: anndata with connectivities
	output: kernel object 

3.2 compute transition matrix
	input: initialized kernel object
	output: kernel object with transition matrix

3.3 combine kernels
	input: 2 kernels and combining ratio (combined_kernel = 0.8 * vk + 0.2 * ck)
	output: combined kernel with transition matrix

4. initialize estimator for matrix decomposition

from cellrank.tl.estimators import GPCCA
g = GPCCA(combined_kernel)

#note best estimator is GPCCA use this by default
input: kernel with transition matrix
output: *estimator* initialized object with transition matrix (anndata)

5. compute matrix decomposition

g.compute_schur(n_components=20)

params: 
		- no of components (usually 20)
		- method: default to krylov for GPCCA

input: *estimator* initialized object with transition matrix (anndata)
output: *estimator* object with transition matrix decomposition (anndata)
#nb this can be plotted

6. compute macrostates

g.compute_macrostates(n_states=3, cluster_key="clusters")

params: 
		- number of macrostates if known, or from inspection of re(gamma) vs eigenvalue index plot
		- clustering info (NB don't understand if this can be run without clustering altogether)

input: *estimator* object with transition matrix decomposition	(anndata)	
output: *estimator* object with computed macrostates (anndata)
#nb this can be plotted

7. identify terminal states from macrostates

options:

	- automatic  
		estimator.compute_terminal_states()
		input: *estimator* object with computed macrostates (anndata)
		output: *estimator* object with terminal states in obs (anndata)
	
	- restrict macrostates to a list of states known beforehand
		estimator.set_terminal_states_from_macrostates()
		input: *estimator* object with computed macrostates (anndata) + list of states
		output: *estimator* object with terminal states in obs (anndata)
	
	- manually set terminal states ignoring macrostates
		estimator.set_terminal_states()
		input: *estimator* object with computed macrostates (anndata) + list of states
		output: *estimator* object with terminal states in obs (anndata)

8. estimate fate probabilties (absorption)

g.compute_absorption_probabilities()

input: *estimator* object with terminal states in obs (anndata)
output: *estimator* object with fate probabilities in obs (anndata)

9. find driver genes (correlate GEX with fate prob)

g.compute_lineage_drivers(lineages="Alpha", return_drivers=True)

input:*estimator* object with fate probabilities in obs (anndata)
output *estimator* object with added .var  (anndata)

10. plotting utilities

input: anndata
output: pngs

- g.plot_spectrum()
- g.plot_macrostates()
- g.plot_lineage_drivers("Alpha", n_genes=5)
- cr.pl.circular_projection(adata, keys="clusters", legend_loc="right")