UMAP Hematopoiesis

[StillJosh]

Hi everyone,

In the tutorial for the hematopoiesis dataset, the data is supplied raw with the exception for the UMAP coordinates. We wanted to kindly ask, if you could supply us with the parameters that were used to generate this UMAP. I know that the UMAP generation is stochastic and that we could therefore not reproduce the exact coordinates, but it would still help us a lot!

Thank you very much!

[Xiaojieqiu]

Thanks for this question. We are in the process of adding a new notebook to demonstrate the generation of the umap, the spliced RNA velocity analyses and the correction of backward RNA velocity flow in our cell paper. But to ease your life, here is the code we created that umap (note that we used the excellent scanpy below but a purely dynamo version will be provided in the pending notebook):

firstly, download the raw data via:
!wget https://www.dropbox.com/s/8m8n6fj8yn1ryjd/hsc_all_combined_all_layers.h5ad?dl=1

Then running the following:

import dynamo as dyn 
import scanpy as sc 
adata_dyn = dyn.read('./hsc_all_combined_all_layers.h5ad')

adata_dyn
adata_dyn.obs['time'] = 3
adata_dyn.obs.loc[adata_dyn.obs.batch == 'JL_10', 'time'] = 5
adata_dyn.obs.time.value_counts()

adata_dyn

adata_dyn.obs_names_make_unique()
adata_filtered2 = adata_dyn.copy()
adata_filtered2.obs["nGenes"] = (adata_filtered2.X > 0).sum(1)
adata_filtered2.obs["UMI"] = (adata_filtered2.X).sum(1)

adata_filtered2.X = adata_filtered2.layers['labeled_TC'].copy()
adata_filtered2.X[adata_filtered2.obs.batch == 'JL_10', :] *= 3/5 
sc.pp.recipe_seurat(adata_filtered2)
# sc.pp.normalize_per_cell(adata_filtered2)
# sc.pp.log1p(adata_filtered2)
# sc.pp.scale(adata_filtered2)                              # scale to unit variance and shift to zero mean
sc.tl.pca(adata_filtered2, n_comps=50)

sc.pp.neighbors(adata_filtered2, n_neighbors=30, random_state=1, n_pcs=10)

sc.tl.leiden(adata_filtered2, resolution=0.5)
sc.tl.umap(adata_filtered2, random_state=1, n_components=3)

sc.pl.umap(adata_filtered2, color=['nGenes', 'UMI'])

adata_filtered2.obs.UMI.median(), adata_filtered2.obs.nGenes.median()

adata_filtered2

# plot results
sc.pl.umap(
    adata_filtered2,
    color=[
        "leiden",
        'nGenes', 
        'UMI',
        'batch',
    ],
    legend_fontsize="large",
    ncols=4,
)

# Vijay_genes = ["GFI1", "RUNX1", "IRF4", "IRF8", "SPI1", 'MPO']

# chen_paper_genes = ['PLEK', 'HBB', 'MPO', 'SPIB', 'CD79A', 'DNTT', 'CD34', 'CD164', 'LMO4', 'PF4', "LYZ", "PLEK", "MPO", "LMO4", "HBB"] + paul_Ery + Er
# sc.pl.umap(
#     adata_filtered2,
#     color=adata_filtered2.var_names.intersection(chen_paper_genes + Vijay_genes + ['leiden']),
#     ncols=4,
# )

adata_filtered2 = adata_filtered2[~ adata_filtered2.obs.leiden.isin(["7"])]

# adata_filtered2.write_h5ad("hsc_adata_filtered2.h5ad")

[StillJosh]

Thank you very much for the quick reply! This is indeed really helpful :)