Plots showing projection of perturbation effect in embedding look the same for all regulons

solvi808 commented 6 months ago

Describe the bug This is not a hard error, but it's an issue with the output of the analysis.

For all the regulon markers including markers that should act in opposite directions the perturbation plots show arrows that always point in the same direction, i.e. are exactly the same. See figures attached of the perturbation-flow-plots of TFs Nfatc1 and Trp63. Nfatc1 and Trp63 in my dataset, and in classic SCENIC, have activity in the opposite clusters, and therefore I expected arrows pointing towards the opposite directions. However it looks almost the exact same as the plot showing the Trp63 perturbation which doesn't make sense. The only difference between the plots seems to be slightly due to the stochastic calculation so I am wondering if somehow, it's always being applied on the same regulon.
The perturbation heatmap looks fine (in this particular sample) , it is only the perturbation featureplot that is same for all TFs
This problem is the same for all my samples, I did three separate analysis on three separate 10x samples.
I am using the basically the exact same code as used in the vignette which was at https://scenicplus.readthedocs.io/en/latest/Perturbation_simulation.html (this vignette with perturbation analysis was removed with the last scenic update).

Anyway I supplied the precomputed perturbed matrix ( perturbed_matrices_dict ) but the issue persisted even if I use the same code as in the SCENIC+ vignette.

My Code:

for TF in tqdm(TFs_of_interest, total = len(TFs_of_interest)):
        _ = plot_perturbation_effect_in_embedding(
                scplus_obj = scplus_obj ,
                reduction_name = 'eRegulons_PCA_gene_based' ,
                perturbed_matrix = perturbed_matrices_dict[TF]['0'] ,
                regressors = regressors ,
                n_cpu = NCPUS ,
               # perturbation = { TF : 0} , 
                variable = clustering_FIELD ,
                color_dictionary = { clustering_FIELD: clusterColors } ,
                genes_to_use = hvg , # hvg DEGs
                save = Path( PCA_perturbation_plots_dir , f'{TF}_perturbation.png' ) ,
                figsize = (10, 10) )

perturbation_featureplot