stacked_barplot

Short Description

sm.pl.stacked_barplot: The function allows users to generate a stacked bar plot of a categorical column. The function can generate the plots using matplotlib and Plotly libraries. Plotly is browser based and so it can be used for interactive data exploration.

Function

stacked_barplot(adata, x_axis='imageid', y_axis='phenotype', subset_xaxis=None, subset_yaxis=None, order_xaxis=None, order_yaxis=None, method='percent', plot_tool='matplotlib', matplotlib_cmap=None, matplotlib_bbox_to_anchor=(1, 1.02), matplotlib_legend_loc=2, return_data=False, **kwargs)

Parameters:

Name	Type	Description	Default
adata		AnnData Object	required
x_axis	string	Column name of the data that need to be plotted in the x-axis.	'imageid'
y_axis	string	Column name of the data that need to be plotted in the y-axis.	'phenotype'
subset_xaxis	list	Subset x-axis before plotting. Pass in a list of categories. eg- subset_xaxis = ['ROI_1', 'ROI_5']	None
subset_yaxis	list	Subset y-axis before plotting. Pass in a list of categories. eg- subset_yaxis = ['Celltype_A', 'Celltype_B']	None
order_xaxis	list	Order the x-axis of the plot as needed. Pass in a list of categories. eg- order_xaxis = ['ROI_5', 'ROI_1'] The default is None and will be plotted based on alphabetic order. Please note that if you change the order, pass all categories, failure to do so will generate NaN's.	None
order_yaxis	list	Order the y-axis of the plot as needed. Pass in a list of categories. eg- order_yaxis = ['Celltype_B', 'Celltype_A'] The default is None and will be plotted based on alphabetic order. Please note that if you change the order, pass all categories, failure to do so will generate NaN's.	None
method	string	Available options: 'percent' and 'absolute'. 1) Use Percent to plot the percent proportion. 2) Use 'absolute' to plot the plot the absolute number.	'percent'
plot_tool	string	Available options: 'matplotlib' and 'plotly'. 1) matplotlib uses the standard python plotting method 2) plotly opens the plot in a local browser. Advantage is to be able to hover over the plot and retreive data for plots with large number of categories.	'matplotlib'
matplotlib_cmap	string	Colormap to select colors from. If string, load colormap with that name from matplotlib.	None
matplotlib_bbox_to_anchor	tuple	Bounding box argument used along with matplotlib_legend_loc to control the legend location when using the matplotlib method.	(1, 1.02)
matplotlib_legend_loc	int	Location of legend used along with matplotlib_bbox_to_anchor to control the legend location when using the matplotlib method.	2
return_data	bool	When True, return the data used for plotting.	False
**kwargs		Additional keyword arguments passed to: 1) Pandas DataFrame.plot() when using the matplotlib method (https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.plot.html#pandas.DataFrame.plot)) 2) Plotly.bar() when using the plotly method (https://plotly.com/python-api-reference/generated/plotly.express.bar.html))	{}

Returns:

Type	Description
	Stacked bar plot. If return_data is set to True also returns a dataframe of the data used for the plot.

    # Plot the absolute number of phenotypes using the matplotlib 
    # tool across differnt ROI's
    # ROI column is `epidermis_roi` and phenotypes are stored under `phenotype`

    sm.pl.stacked_barplot (adata,x_axis='epidermis_roi',y_axis='phenotype',
                     method='absolute',plot_tool='matplotlib',
                     figsize=(10, 10))

    # Plot the number of cells normalized to 100% using the plotly 
    # tool across differnt ROI's

    sm.pl.stacked_barplot (adata,x_axis='epidermis_roi',y_axis='phenotype',
                     method='percent',plot_tool='plotly',
                     color_discrete_sequence=px.colors.qualitative.Alphabet)

    # Same as above but visualize only a subset of ROI's and a subset of 
    # phenotypes
    subset_xaxis = ['epidermis_1', 'epidermis_5', 'epidermis_6']
    subset_yaxis = ['APCs', 'Keratinocytes', 'Macrophages']

    sm.pl.stacked_barplot (adata,x_axis='epidermis_roi',y_axis='phenotype',
                            subset_xaxis=subset_xaxis,subset_yaxis=subset_yaxis,
                            method='percent',plot_tool='plotly')

    # Visualize absolute number of phenotypes and return the data into a 
    # dataframe `absolute_number`
    absolute_number = sm.pl.stacked_barplot (adata,x_axis='epidermis_roi',
                      y_axis='phenotype', method='absolute',
                      plot_tool='matplotlib', return_data=True)

Source code in scimap/plotting/_stacked_barplot.py

def stacked_barplot (adata, x_axis='imageid', y_axis='phenotype', subset_xaxis=None, subset_yaxis=None, 
                     order_xaxis=None, order_yaxis=None,
                     method='percent', plot_tool='matplotlib', matplotlib_cmap=None, 
                     matplotlib_bbox_to_anchor=(1,1.02), matplotlib_legend_loc=2, 
                     return_data=False, **kwargs):
    """
Parameters:
    adata : AnnData Object

    x_axis (string):  
        Column name of the data that need to be plotted in the x-axis.

    y_axis (string):  
        Column name of the data that need to be plotted in the y-axis.

    subset_xaxis (list):  
        Subset x-axis before plotting. Pass in a list of categories. eg- subset_xaxis = ['ROI_1', 'ROI_5']

    subset_yaxis (list):  
        Subset y-axis before plotting. Pass in a list of categories. eg- subset_yaxis = ['Celltype_A', 'Celltype_B']

    order_xaxis (list):  
        Order the x-axis of the plot as needed. Pass in a list of categories. eg- order_xaxis = ['ROI_5', 'ROI_1']
        The default is None and will be plotted based on alphabetic order. Please note that if you change the order, pass all categories, failure to do so
        will generate NaN's.

    order_yaxis (list):  
        Order the y-axis of the plot as needed. Pass in a list of categories. eg- order_yaxis = ['Celltype_B', 'Celltype_A']
        The default is None and will be plotted based on alphabetic order. Please note that if you change the order, pass all categories, failure to do so
        will generate NaN's.

    method (string):  
        Available options: 'percent' and 'absolute'. 
        1) Use Percent to plot the percent proportion.  
        2) Use 'absolute' to plot the plot the absolute number.  

    plot_tool (string):  
        Available options: 'matplotlib' and 'plotly'.  
        1) matplotlib uses the standard python plotting method  
        2) plotly opens the plot in a local browser. Advantage is to be able   
        to hover over the plot and retreive data for plots with large number of categories.

    matplotlib_cmap (string):  
        Colormap to select colors from. If string, load colormap with that name from matplotlib. 

    matplotlib_bbox_to_anchor (tuple):  
        Bounding box argument used along with matplotlib_legend_loc to control
        the legend location when using the matplotlib method.

    matplotlib_legend_loc (int):  
        Location of legend used along with matplotlib_bbox_to_anchor to control
        the legend location when using the matplotlib method.

    return_data (bool):  
        When True, return the data used for plotting.

    **kwargs : Additional keyword arguments passed to:  
        1) Pandas DataFrame.plot() when using the `matplotlib` method (https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.plot.html#pandas.DataFrame.plot))  
        2) Plotly.bar() when using the `plotly` method (https://plotly.com/python-api-reference/generated/plotly.express.bar.html))  

Returns:  
    Stacked bar plot. If return_data is set to `True` also returns a dataframe of the data used for the plot.


Example:
```python
    # Plot the absolute number of phenotypes using the matplotlib 
    # tool across differnt ROI's
    # ROI column is `epidermis_roi` and phenotypes are stored under `phenotype`

    sm.pl.stacked_barplot (adata,x_axis='epidermis_roi',y_axis='phenotype',
                     method='absolute',plot_tool='matplotlib',
                     figsize=(10, 10))

    # Plot the number of cells normalized to 100% using the plotly 
    # tool across differnt ROI's

    sm.pl.stacked_barplot (adata,x_axis='epidermis_roi',y_axis='phenotype',
                     method='percent',plot_tool='plotly',
                     color_discrete_sequence=px.colors.qualitative.Alphabet)

    # Same as above but visualize only a subset of ROI's and a subset of 
    # phenotypes
    subset_xaxis = ['epidermis_1', 'epidermis_5', 'epidermis_6']
    subset_yaxis = ['APCs', 'Keratinocytes', 'Macrophages']

    sm.pl.stacked_barplot (adata,x_axis='epidermis_roi',y_axis='phenotype',
                            subset_xaxis=subset_xaxis,subset_yaxis=subset_yaxis,
                            method='percent',plot_tool='plotly')

    # Visualize absolute number of phenotypes and return the data into a 
    # dataframe `absolute_number`
    absolute_number = sm.pl.stacked_barplot (adata,x_axis='epidermis_roi',
                      y_axis='phenotype', method='absolute',
                      plot_tool='matplotlib', return_data=True)

```
    """


    # create the dataframe with details
    data = pd.DataFrame(adata.obs)[[x_axis,y_axis]].astype(str)

    # subset the data if needed
    #if subset_data is not None:data = data[data[list(subset_data.keys())[0]].isin(list(subset_data.values())[0])]

    if subset_xaxis is not None:
        if isinstance(subset_xaxis, str):
            subset_xaxis = [subset_xaxis]
        data = data[data[x_axis].isin(subset_xaxis)]
    if subset_yaxis is not None:
        if isinstance(subset_yaxis, str):
            subset_yaxis = [subset_yaxis]
        data = data[data[y_axis].isin(subset_yaxis)]


    # Method: Absolute or Percentile
    if method == 'percent':
        total = data.groupby([x_axis,y_axis]).size().unstack().fillna(0).sum(axis=1)
        rg = pd.DataFrame(data.groupby([x_axis,y_axis]).size().unstack().fillna(0).div(total, axis=0).stack())
    elif method == 'absolute':
        rg = pd.DataFrame(data.groupby([x_axis,y_axis]).size().unstack().fillna(0).stack())
    else:
        raise ValueError('method should be either percent or absolute')

    # change column name
    rg.columns = ['count']

    # Add the index as columns in the data frame    
    rg.reset_index(inplace=True)  

    # re-order the x oy y axis if requested by user
    if order_xaxis is not None:
        rg[x_axis] = rg[x_axis].astype('category')
        rg[x_axis] = rg[x_axis].cat.reorder_categories(order_xaxis)
        rg = rg.sort_values(x_axis)
    if order_yaxis is not None:
        rg[y_axis] = rg[y_axis].astype('category')
        rg[y_axis] = rg[y_axis].cat.reorder_categories(order_yaxis)
        rg = rg.sort_values(y_axis)
    if order_xaxis and order_yaxis is not None:
        rg = rg.sort_values([x_axis, y_axis])

    pivot_df = rg.pivot(index=x_axis, columns=y_axis, values='count')

    # Plotting tool
    if plot_tool == 'matplotlib':

        if matplotlib_cmap is None:
            if len(rg[y_axis].unique()) <= 9:
                matplotlib_cmap = "Set1"        
            elif len(rg[y_axis].unique()) > 9 and len(rg[y_axis].unique()) <=20:
                matplotlib_cmap = plt.cm.tab20      #tab20  
            else:
                matplotlib_cmap = plt.cm.gist_ncar

        # Plotting
        # add width if not passed via parameters
        try:
            width
        except NameError:
            width=0.9
        # actual plotting   
        p = pivot_df.plot.bar(stacked=True, cmap=matplotlib_cmap, width=width,  **kwargs)
        handles, labels = p.get_legend_handles_labels() # for reversing the order of the legend
        p.legend(reversed(handles), reversed(labels), bbox_to_anchor=matplotlib_bbox_to_anchor, loc=matplotlib_legend_loc)

    elif plot_tool == 'plotly':

        fig = px.bar(rg, x=x_axis, y="count", color=y_axis, **kwargs)
        fig.update_layout({'plot_bgcolor': 'rgba(0, 0, 0, 0)',
                           'paper_bgcolor': 'rgba(0, 0, 0, 0)'},
                          xaxis = dict(tickmode='linear') #type = 'category'
                          )
        fig.show()


    else:

        raise ValueError('plot_tool should be either matplotlib or plotly')

    # Return data
    if return_data is True:
        return pivot_df