Scanpy compatibility

Since Swan now uses the AnnData data structure to track abundance for transcripts, TSSs, TESs, and edges, this makes these objects directly compatible with the amazing suite of Scanpy preprocessing, analysis, and visualization tools.

For this brief overview of some cool things you can do with Swan, I'll be using the Swan object of the data from my lab's 2021 preprint on LR-Split-seq to showcase how specifically Scanpy can be used for single-cell data.

import swan_vis as swan
import scanpy as sc

fname = '/Users/fairliereese/mortazavi_lab/data/c2c12_paper_2020/sc_pacbio/210618/c2c12.p'
sg = swan.read(fname)

Read in graph from /Users/fairliereese/mortazavi_lab/data/c2c12_paper_2020/sc_pacbio/210618/c2c12.p

UMAP

Provide the corresponding AnnData from the SwanGraph and the name of the transcript / edge / TES / TSS you want to plot.

sc.pl.umap(sg.tss_adata,
           color='Nisch_23',
           gene_symbols='tss_name',
           layer='tpm',
           frameon=False,
           size=120,
           show=False,
           cmap='magma')

<AxesSubplot:title={'center':'Nisch_23'}, xlabel='UMAP1', ylabel='UMAP2'>

png

# add transcript name to sg.adata really quick
sg.adata.var = sg.adata.var.merge(sg.t_df, left_index=True, right_index=True)

sc.pl.umap(sg.adata,
           color='Tpm2-202',
           gene_symbols='tname',
           frameon=False,
           size=120,
           show=False,
           layer='tpm',
           cmap='viridis')

... storing 'ill_umi_count' as categorical
... storing 'ill_gene_count' as categorical
... storing 'bc3' as categorical
... storing 'bc2' as categorical
... storing 'bc1' as categorical
... storing 'well' as categorical
... storing 'primer_type' as categorical
... storing 'raw_bc1' as categorical
... storing 'gid' as categorical
... storing 'gname' as categorical
... storing 'novelty' as categorical





<AxesSubplot:title={'center':'Tpm2-202'}, xlabel='UMAP1', ylabel='UMAP2'>

png

sc.pl.umap(sg.tes_adata,
           color='Tnnt2_18',
           gene_symbols='tes_name',
           layer='tpm',
           frameon=False,
           size=120,
           show=False,
           cmap='magma')

<AxesSubplot:title={'center':'Tnnt2_18'}, xlabel='UMAP1', ylabel='UMAP2'>

png

Dotplots, matrixplots, and heatmaps

Scanpy provides many different options to visualize expression of multiple variables per group. Some of my favorites are dotplots, matrixplots, and heatmaps.

Here, I'll plot the expression of each Tpm2 isoform per cluster using a dotplot. The cluster column name in sg.adata.obs is called leiden.

sg.adata.obs.leiden.head()

index
TGGAACAAGTGTTCTACGTTCGAG    3
GTACGCAATCCGTCTAACGCCGGC    2
AAGAGATCCCGTGAGAACGCCGGC    1
ACTATGCAACACGACCGTGCTAGC    5
CAATGGAACTAAGGTCGTTCAACA    1
Name: leiden, dtype: category
Categories (7, object): ['1', '2', '3', '4', '5', '6', '7']

tpm2_isos = sg.t_df.loc[sg.t_df.gname == 'Tpm2', 'tname'].tolist()
tpm2_isos

['ENCODEMT000193235',
 'ENCODEMT000193244',
 'ENCODEMT000193257',
 'ENCODEMT000193335',
 'ENCODEMT000193446',
 'ENCODEMT000193549',
 'ENCODEMT000193650',
 'ENCODEMT000193696',
 'ENCODEMT000193732',
 'ENCODEMT000193749',
 'ENCODEMT000193807',
 'ENCODEMT000193825',
 'ENCODEMT000194010',
 'ENCODEMT000194052',
 'ENCODEMT000194060',
 'ENCODEMT000194110',
 'ENCODEMT000194146',
 'ENCODEMT000194260',
 'ENCODEMT000194282',
 'ENCODEMT000194549',
 'ENCODEMT000194721',
 'ENCODEMT000194729',
 'ENCODEMT000194759',
 'ENCODEMT000194779',
 'ENCODEMT000194829',
 'Tpm2-201',
 'Tpm2-202',
 'Tpm2-203',
 'Tpm2-204',
 'Tpm2-205',
 'Tpm2-206']

# merge with transcript names and annotation information
sg.adata.var = sg.adata.var.merge(sg.t_df[['tname', 'annotation']], left_index=True, right_index=True)
sg.adata.var.head()

	tid	tname	annotation
tid
ENCODEMT000141768	ENCODEMT000141768	ENCODEMT000141768	False
ENCODEMT000141769	ENCODEMT000141769	ENCODEMT000141769	False
ENCODEMT000141770	ENCODEMT000141770	ENCODEMT000141770	False
ENCODEMT000141771	ENCODEMT000141771	ENCODEMT000141771	False
ENCODEMT000141772	ENCODEMT000141772	ENCODEMT000141772	False

sc.pl.dotplot(sg.adata,
              var_names=tpm2_isos,
              layer='tpm',
              groupby='leiden',
              gene_symbols='tname')

png

Here, I'll use a heatmap to show the expression of each known isoform of Tpm2 in each cluster.

# limit to just the annotated isoforms
tpm2_known_isos = sg.t_df.loc[(sg.t_df.gname == 'Tpm2')&(sg.t_df.annotation == True), 'tname'].tolist()
tpm2_known_isos

['Tpm2-201', 'Tpm2-202', 'Tpm2-203', 'Tpm2-204', 'Tpm2-205', 'Tpm2-206']

sc.pl.heatmap(sg.adata,
              var_names=tpm2_known_isos,
              layer='tpm',
              groupby='leiden',
              gene_symbols='tname')

png

I can also plot using different metadata to group the cells by, like the input sample:

sg.adata.obs['sample'].head()

index
TGGAACAAGTGTTCTACGTTCGAG     MB_cells
GTACGCAATCCGTCTAACGCCGGC    MB_nuclei
AAGAGATCCCGTGAGAACGCCGGC    MB_nuclei
ACTATGCAACACGACCGTGCTAGC    MT_nuclei
CAATGGAACTAAGGTCGTTCAACA    MB_nuclei
Name: sample, dtype: category
Categories (3, object): ['MB_cells', 'MB_nuclei', 'MT_nuclei']

mp = sc.pl.heatmap(sg.adata,
              var_names=tpm2_known_isos,
              layer='tpm',
              groupby='sample',
              gene_symbols='tname')

png

Here's the same data plotted in a matrix plot, which aggregates expression across each category (but doesn't show the colors for each category).

mp = sc.pl.matrixplot(sg.adata,
              var_names=tpm2_known_isos,
              layer='tpm',
              groupby='sample',
              gene_symbols='tname')

png

Of course, you can make all the above plots with TSS or TES expression data as well! Such as the dotplot below for different TSSs of Nisch.

nisch_tss = sg.tss_adata.var.loc[sg.tss_adata.var.gname == 'Nisch', 'tss_name'].tolist()
nisch_tss = ['Nisch_12', 'Nisch_23', 'Nisch_36']
sc.pl.dotplot(sg.tss_adata,
              var_names=nisch_tss,
              layer='tpm',
              groupby='leiden',
              gene_symbols='tss_name')

png

Violin plots

Scanpy also offers violin plots which are often used to visualize expression distributions of different categories.

First, I'll show the expression of a single isoform across the different identified cell types:

tname = 'Tpm2-202'
tid = sg.adata.var.loc[sg.adata.var.tname == tname, 'tid'].values[0]
sc.pl.violin(sg.adata,
             keys=tid,
             groupby='celltype',
             layer='tpm',
             ylabel=tname)

png

Of course you can also plot the expression of TSSs and TESs from your SwanGraph using this strategy as well, as shown below.

tes_name = 'Myog_2'
tss_id = sg.tes_adata.var.loc[sg.tes_adata.var.tes_name == tes_name].index.values[0]
sc.pl.violin(sg.tes_adata,
             keys=tss_id,
             groupby='leiden',
             layer='tpm',
             ylabel=tes_name)

png