tosches turtle data

In this tutorial, we demonstrate how to use DTNE to analyze a single cell dataset, which comes from Tosches,et al.(2018) and can be downloaded with the link.

Load library and data.

import numpy as np
import scanpy as sc
import matplotlib.pyplot as plt

adata = sc.read_h5ad('Tosches_turtle.h5ad')
adata

AnnData object with n_obs × n_vars = 18664 × 23500
    obs: 'sample', 'region', 'donor', 'cell_type1', 'gender', 'organism', 'dataset_name', 'platform', 'organ', 'data_type', 'cell_ontology_class', 'cell_ontology_id', 'n_genes', 'n_counts', '__libsize__'
    var: 'variable_genes'
    uns: 'author_genes', 'cell_ontology_class_colors', 'cell_type1_colors', 'data_quality', 'donor_colors', 'gender_colors', 'neighbors', 'region_colors', 'sample_colors', 'umap'
    obsm: 'X_umap', 'latent'
    obsp: 'connectivities', 'distances'

cell_ontology_class = adata.obs['cell_ontology_class']
turtle_cluster_set = set(cell_ontology_class)
n_groups = len(turtle_cluster_set)
turtle_zip_types = zip(sorted(turtle_cluster_set),range(len(turtle_cluster_set)))
turtle_dict_types = dict(turtle_zip_types)
turtle_cell_clusters  =  [turtle_dict_types[i] for i in cell_ontology_class]

use DTNE for visualization and clustering.

from dtne import *

X = adata.obsm['latent']
dtne_operator = DTNE(k_neighbors = 30)
Y = dtne_operator.fit_transform(X)
dtne_cluster = dtne_operator.cluster_cells(n_clusters=n_groups)
plt.scatter(Y[:,0], Y[:,1], c=dtne_cluster,s=1,cmap='viridis')

<matplotlib.collections.PathCollection at 0x1a0ad0c1710>

dtne_operator = DTNE(k_neighbors = 30, min_dist = 0.9, solver = 'umap')
Y2 = dtne_operator.fit_transform(X)
plt.scatter(Y2[:,0], Y2[:,1], c=dtne_cluster,s=1,cmap='viridis')

<matplotlib.collections.PathCollection at 0x1a0abe3a2d0>

Visualize different clustering results using UMAP

X_umap = adata.obsm['X_umap']

sc.tl.leiden(adata, resolution=0.068, use_weights=True,) # flavor='leidenalg'
leiden_group = np.array(adata.obs['leiden'].values).astype('int')

C:\Users\jiang\AppData\Local\Temp\ipykernel_11696\2219162274.py:1: FutureWarning: In the future, the default backend for leiden will be igraph instead of leidenalg.

 To achieve the future defaults please pass: flavor="igraph" and n_iterations=2.  directed must also be False to work with igraph's implementation.
  sc.tl.leiden(adata, resolution=0.068, use_weights=True,) # flavor='leidenalg'

fig,axes = plt.subplots(ncols=3,figsize=(12, 4))
for ax in axes.ravel():
    ax.set_axis_off()

title_names = ['default cluster','leiden cluster','dtne cluster',]
cell_colors = [turtle_cell_clusters,leiden_group,dtne_cluster]

for i,ax in enumerate(axes):
    ax.set_title(title_names[i], fontsize=10)
    ax.scatter(X_umap[:,0], X_umap[:,1],c=cell_colors[i],s=1,cmap='viridis')

ic_ids1 = [1 if turtle_cell_clusters[i] == 2 else 0 for i in range(len(turtle_cell_clusters))]
ic_ids2 = [1 if leiden_group[i] == 1 else 0 for i in range(len(leiden_group))]

ic_ids31 = [1 if dtne_cluster[i] == 0 else 0 for i in range(len(dtne_cluster))]
ic_ids32 = [0.5 if dtne_cluster[i] == 6 else 0 for i in range(len(dtne_cluster))]
ic_ids3 = np.array(ic_ids31) + np.array(ic_ids32)

fig,axes = plt.subplots(ncols=3,figsize=(12, 4))
for ax in axes.ravel():
    ax.set_axis_off()

title_names = ['default cluster','leiden cluster','dtne cluster',]
cell_colors = [turtle_cell_clusters,leiden_group,dtne_cluster]
ic_ids = [ic_ids1,ic_ids2,ic_ids3]

for i,ax in enumerate(axes):
    ax.set_title(title_names[i], fontsize=10)
    ax.scatter(X_umap[:,0], X_umap[:,1],c=cell_colors[i],s=1,cmap='viridis')
    ax.scatter(X_umap[:,0], X_umap[:,1],c=ic_ids[i],s=5,alpha=0.5,cmap='Grays')

UMAP prioritize showing similarities between nearby points but may distort relationships between distant points. The UMAP visualizations might not always align perfectly with the actual clustering results.