In [1]:

%config InlineBackend.figure_format = 'retina'
%pylab inline

from glob import glob

import anndata
import pandas as pd

Populating the interactive namespace from numpy and matplotlib

In [2]:

datasets = [
    anndata.read('../Data/output/10x v3 HEK293T.h5ad'),
    anndata.read('../Data/output/10x v3 NIH3T3.h5ad'),
    anndata.read('../Data/output/Klein et al 2015.h5ad'),
    anndata.read('../Data/output/Macosko et al 2015.h5ad'),
    anndata.read('../Data/output/Svensson et al 2017 (1).h5ad'),
    anndata.read('../Data/output/Svensson et al 2017 (2).h5ad'),
    anndata.read('../Data/output/Zheng et al 2017.h5ad'),
    anndata.read('../Data/output/Padovan-Merhar et al 2015 (SMARTer).h5ad'),
    anndata.read('../Data/output/10x v3 PBMC.h5ad')
]

In [7]:

annotation = {
    'Macosko et al 2015': 'Solution (ERCC spike-ins)',
    'Svensson et al 2017 (1)': 'Solution (brain endogenuous RNA & ERCC spike-ins)',
    'Zheng et al 2017': 'Solution (ERCC spike-ins)',
    'Svensson et al 2017 (2)': 'Solution (brain endogenuous RNA & ERCC spike-ins)',
    '10x v3 HEK293T': 'Single cells (homogenuous)',
    '10x v3 NIH3T3': 'Single cells (homogenuous)',
    '10x v3 PBMC': 'Single cells (heterogenuous)',
    'Klein et al 2015': 'Solution (K562 endogenuous RNA & ERCC spike-ins)',
    'Padovan-Merhar et al 2015 (SMARTer)': 'Single cells (homogeneuous, full length)'
}

In [11]:

location = {
    'Macosko et al 2015': 6,
    'Svensson et al 2017 (1)': 3,
    'Zheng et al 2017': 7,
    'Svensson et al 2017 (2)': 5,
    '10x v3 HEK293T': 2,
    '10x v3 NIH3T3': 1,
    '10x v3 PBMC': 0,
    'Klein et al 2015': 4,
    'Padovan-Merhar et al 2015 (SMARTer)': 8
}

In [12]:

mins = []
maxs = []
for adata in datasets:
    difference1 = adata.var['empirical_zero_fraction'] - adata.var['global_zero_fraction']
    mins.append(difference1.min()),
    maxs.append(difference1.max())

min(mins), max(maxs)

Out[12]:

(-0.3979434284180565, 0.8805231468904677)

In [13]:

fig = plt.figure(figsize=(15, 18))

outer_grid = fig.add_gridspec(3, 3, hspace=0.4, wspace=0.6)

for adata in datasets:
    
    i = location[adata.uns['name']]
    grid_box = outer_grid[i]
    
    inner_grid = grid_box.subgridspec(2, 1)
    
    # -- Global -- 
    
    ax = fig.add_subplot(inner_grid[0])
    
    ax.set_title('Poisson')
    
    ax.set_xscale('log')
    ax.set_xlim(left=1e-9, right=1e-1)
    
    ax.scatter(adata.var['scaled_count_mean'],
               adata.var['empirical_zero_fraction'],
               c='k', label='Observed', rasterized=True);
    ax.scatter(adata.var['scaled_count_mean'],
               adata.var['poisson_zero_fraction'],
               ec='w', c='grey', label='Expected', rasterized=True);
    
    ax.set_ylabel('Fraction zeros')
    
    ax.legend(title='Genes', loc='lower left', scatterpoints=3, fontsize=8)
    
    ax.spines['top'].set_visible(False)
    ax.spines['right'].set_visible(False)
    
    ## _
    
    ax = fig.add_subplot(inner_grid[1])
    
    ax.set_xscale('log')
    ax.set_xlim(left=1e-9, right=1e-1)
    ax.set_ylim(top=1., bottom=-0.4)
    
    difference = adata.var['empirical_zero_fraction'] - adata.var['poisson_zero_fraction']
    ax.scatter(adata.var['scaled_count_mean'],
               difference,
               c='k', marker='.', label='Genes', rasterized=True)
    
    ax.set_ylabel('Difference \n(Observed - Expected)')
    ax.set_xlabel('Mean (of scaled counts)')
    
    ax.legend(loc='lower left', scatterpoints=3)
    
    ax.spines['top'].set_visible(False)
    ax.spines['right'].set_visible(False)
    
    
    # -- Annotation --
    
    bbox = grid_box.get_position(fig)
    x = (bbox.x0 + bbox.x1) / 2
    y = bbox.y1 + 0.02
    
    title_str = adata.uns['name'] + '\n' + annotation[adata.uns['name']]
    fig.text(x, y, title_str, ha='center', fontsize=12)
    
fig.savefig('../Figures/poisson_plot.pdf', dpi=500, bbox_inches='tight')

In [ ]: