umap

Short Description

sm.pl.umap: This function facilitates the creation of scatter plots based on UMAP (Uniform Manifold Approximation and Projection) embeddings stored in an AnnData object. It offers extensive customization options for visualizing high-dimensional data reduced to two dimensions, including the ability to color points by gene expression levels, metadata annotations, or other categorical or continuous variables. Users can leverage this function to explore and interpret complex datasets visually, enhancing the understanding of underlying biological variations and relationships.

Function

umap(adata, color=None, layer=None, use_raw=False, log=False, label='umap', cmap='vlag', palette=None, alpha=0.8, figsize=(5, 5), s=None, ncols=None, tight_layout=False, return_data=False, save_figure=None, **kwargs)

Parameters:

Name	Type	Description	Default
adata	AnnData	The annotated data matrix.	required
color	list	List of keys from adata.obs.columns or adata.var.index to color the plot. Allows multiple keys for facetted plotting.	None
layer	str	Specifies the AnnData layer to use for UMAP calculations. Defaults to using adata.X.	None
use_raw	bool	If True, uses adata.raw.X for coloring the plot, useful for visualizing gene expression on UMAP.	False
log	bool	Applies log transformation (np.log1p) to the data before plotting. Useful for gene expression data.	False
label	str	Key in adata.obsm where UMAP coordinates are stored.	'umap'
cmap	str	Colormap for continuous variables. Supports matplotlib colormap names and objects.	'vlag'
palette	dict	Specific colors for different categories as a dictionary mapping from categories to colors.	None
alpha	float	Transparency level of the points. Ranges from 0 (transparent) to 1 (opaque).	0.8
figsize	tuple	Figure size specified as (width, height) in inches.	(5, 5)
s	int	Size of the points in the plot.	None
ncols	int	Number of columns for facetted plotting.	None
tight_layout	bool	Adjusts subplot params for a tight layout.	False
return_data	bool	If True, returns the DataFrame containing data used for plotting instead of displaying the plot.	False
save_figure	str	Path and filename to save the figure. File extension determines the format (e.g., .pdf, .png).	None
**kwargs		Additional keyword arguments passed to matplotlib plot function.	{}

Returns:

Name	Type	Description
Plot	matplotlib	Optionally returns the data used for plotting if return_data=True.

Example

# Basic UMAP visualization with default settings
sm.pl.umap(adata, color='cell_type')

# UMAP visualization with log transformation and custom colormap
sm.pl.umap(adata, color='gene_expression', log=True, cmap='coolwarm')

# Facetted UMAP plotting with custom point size and saved figure
sm.pl.umap(adata, color=['cell_type', 'condition'], s=100, figsize=(10, 5), save_figure='/path/to/umap_plot.png')

Source code in scimap/plotting/umap.py

def umap (adata, 
          color=None, 
          layer=None, 
          use_raw=False, 
          log=False, 
          label='umap',
          cmap='vlag', 
          palette=None, 
          alpha=0.8, 
          figsize=(5, 5), 
          s=None, 
          ncols=None, 
          tight_layout=False, 
          return_data=False, 
          save_figure=None, **kwargs):
    """
Parameters:
        adata (anndata.AnnData):  
            The annotated data matrix.

        color (list, optional):  
            List of keys from `adata.obs.columns` or `adata.var.index` to color the plot. 
            Allows multiple keys for facetted plotting.

        layer (str, optional):  
            Specifies the AnnData layer to use for UMAP calculations. Defaults to using `adata.X`.

        use_raw (bool, optional):  
            If True, uses `adata.raw.X` for coloring the plot, useful for visualizing gene expression on UMAP.

        log (bool, optional):  
            Applies log transformation (`np.log1p`) to the data before plotting. Useful for gene expression data.

        label (str, optional):  
            Key in `adata.obsm` where UMAP coordinates are stored.

        cmap (str, optional):  
            Colormap for continuous variables. Supports matplotlib colormap names and objects.

        palette (dict, optional):  
            Specific colors for different categories as a dictionary mapping from categories to colors.

        alpha (float, optional):  
            Transparency level of the points. Ranges from 0 (transparent) to 1 (opaque).

        figsize (tuple, optional):  
            Figure size specified as (width, height) in inches.

        s (int, optional):  
            Size of the points in the plot.

        ncols (int, optional):  
            Number of columns for facetted plotting.

        tight_layout (bool, optional):  
            Adjusts subplot params for a tight layout.

        return_data (bool, optional):  
            If True, returns the DataFrame containing data used for plotting instead of displaying the plot.

        save_figure (str, optional):  
            Path and filename to save the figure. File extension determines the format (e.g., `.pdf`, `.png`).

        **kwargs:  
            Additional keyword arguments passed to matplotlib plot function.

Returns:
        Plot (matplotlib):
                Optionally returns the data used for plotting if `return_data=True`.

Example:
    ```python

    # Basic UMAP visualization with default settings
    sm.pl.umap(adata, color='cell_type')

    # UMAP visualization with log transformation and custom colormap
    sm.pl.umap(adata, color='gene_expression', log=True, cmap='coolwarm')

    # Facetted UMAP plotting with custom point size and saved figure
    sm.pl.umap(adata, color=['cell_type', 'condition'], s=100, figsize=(10, 5), save_figure='/path/to/umap_plot.png')

    ```
    """

    # check if umap tool has been run
    try:
        adata.obsm[label]
    except KeyError:
        raise KeyError("Please run `sm.tl.umap(adata)` first")

    # identify the coordinates
    umap_coordinates = pd.DataFrame(adata.obsm[label],index=adata.obs.index, columns=['umap-1','umap-2'])

    # other data that the user requests
    if color is not None:
        if isinstance(color, str):
            color = [color]
        # identify if all elemets of color are available        
        if set(color).issubset(list(adata.var.index) + list(adata.obs.columns)) is False:
            raise ValueError("Element passed to `color` is not found in adata, please check!")

        # organise the data
        if any(item in color for item in list(adata.obs.columns)):
            adataobs = adata.obs.loc[:, adata.obs.columns.isin(color)]
            adataobs = adataobs.apply(lambda x: x.astype('category'))

        else:
            adataobs = None

        if any(item in color for item in list(adata.var.index)):
            # find the index of the marker
            marker_index = np.where(np.isin(list(adata.var.index), color))[0]
            if layer is not None:
                adatavar = adata.layers[layer][:, np.r_[marker_index]]
            elif use_raw is True:
                adatavar = adata.raw.X[:, np.r_[marker_index]]
            else:
                adatavar = adata.X[:, np.r_[marker_index]]
            adatavar = pd.DataFrame(adatavar, index=adata.obs.index, columns = list(adata.var.index[marker_index]))
        else:
            adatavar = None

        # combine all color data
        if adataobs is not None and adatavar is not None:
            color_data = pd.concat ([adataobs, adatavar], axis=1)
        elif adataobs is not None and adatavar is None:
            color_data = adataobs
        elif adataobs is None and adatavar is not None:
            color_data = adatavar
    else:
        color_data = None

    # combine color data with umap coordinates
    if color_data is not None:
        final_data = pd.concat([umap_coordinates, color_data], axis=1)
    else:
        final_data = umap_coordinates

    # create some reasonable defaults
    # estimate number of columns in subpolt
    nplots = len(final_data.columns) - 2 # total number of plots
    if ncols is None:
        if nplots >= 4:
            subplot = [math.ceil(nplots / 4), 4]
        elif nplots == 0:
            subplot = [1, 1]
        else:
            subplot = [math.ceil(nplots / nplots), nplots]
    else:
        subplot = [math.ceil(nplots /ncols), ncols]

    if nplots == 0:
        n_plots_to_remove = 0
    else:
        n_plots_to_remove = np.prod(subplot) - nplots # figure if we have to remove any subplots

    # size of points
    if s is None:
        if nplots == 0:
            s = (100000 / adata.shape[0])
        else:
            s = (100000 / adata.shape[0]) / nplots


    # if there are categorical data then assign colors to them
    if final_data.select_dtypes(exclude=["number","bool_","object_"]).shape[1] > 0:
        # find all categories in the dataframe
        cat_data = final_data.select_dtypes(exclude=["number","bool_","object_"])
        # find all categories
        all_cat = []
        for i in cat_data.columns:
            all_cat.append(list(cat_data[i].cat.categories))

        # generate colormapping for all categories
        less_9 = [colors.rgb2hex(x) for x in sns.color_palette('Set1')]
        nineto20 = [colors.rgb2hex(x) for x in sns.color_palette('tab20')]
        greater20 = [colors.rgb2hex(x) for x in sns.color_palette('gist_ncar', max([len(i) for i in all_cat]))]

        all_cat_colormap = dict()
        for i in range(len(all_cat)):
            if len(all_cat[i]) <= 9:
                dict1 = dict(zip(all_cat[i] , less_9[ : len(all_cat[i]) ]   ))
            elif len(all_cat[i]) > 9 and len(all_cat[i]) <= 20:
                dict1 = dict(zip(all_cat[i] , nineto20[ : len(all_cat[i]) ]   ))
            else:
                dict1 = dict(zip(all_cat[i] , greater20[ : len(all_cat[i]) ]   ))
            all_cat_colormap.update(dict1)

        # if user has passed in custom colours update the colors
        if palette is not None:
            all_cat_colormap.update(palette)
    else:
        all_cat_colormap = None


    # plot
    fig, ax = plt.subplots(subplot[0],subplot[1], figsize=figsize)
    plt.rcdefaults()
    #plt.rcParams['axes.facecolor'] = 'white'

    # remove unwanted axes
    #fig.delaxes(ax[-1])
    if n_plots_to_remove > 0:
        for i in range(n_plots_to_remove):
            fig.delaxes(ax[-1][  (len(ax[-1])-1)-i :   (len(ax[-1]))-i   ][0])

    # to make sure the ax is always 2x2
    if any(i > 1 for i in subplot):
        if any(i == 1 for i in subplot):
            ax = ax.reshape(subplot[0],subplot[1])

    if nplots == 0:
        ax.scatter(x = final_data['umap-1'], y = final_data['umap-2'], s=s, cmap=cmap, alpha=alpha, **kwargs)
        plt.xlabel("UMAP-1"); plt.ylabel("UMAP-2") 
        plt.tick_params(right= False,top= False,left= False, bottom= False)
        ax.get_xaxis().set_ticks([]); ax.get_yaxis().set_ticks([])
        if tight_layout is True:
            plt.tight_layout()

    elif all(i == 1 for i in subplot):
        column_to_plot = [e for e in list(final_data.columns) if e not in ('umap-1', 'umap-2')][0]
        if all_cat_colormap is None:
            im = ax.scatter(x = final_data['umap-1'], y = final_data['umap-2'], s=s, 
                       c=final_data[column_to_plot],
                       cmap=cmap, alpha=alpha, **kwargs)
            plt.colorbar(im, ax=ax)
        else:
            ax.scatter(x = final_data['umap-1'], y = final_data['umap-2'], s=s, 
                       c=final_data[column_to_plot].map(all_cat_colormap),
                       cmap=cmap, alpha=alpha, **kwargs)
            # create legend
            patchList = []
            for key in list(final_data[column_to_plot].unique()):
                data_key = mpatches.Patch(color=all_cat_colormap[key], label=key)
                patchList.append(data_key)    
                ax.legend(handles=patchList,bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)

        plt.xlabel("UMAP-1"); plt.ylabel("UMAP-2") 
        plt.title(column_to_plot)
        plt.tick_params(right= False,top= False,left= False, bottom= False)
        ax.set(xticklabels = ([])); ax.set(yticklabels = ([]))
        if tight_layout is True:
            plt.tight_layout()

    else: 
        column_to_plot = [e for e in list(final_data.columns) if e not in ('umap-1', 'umap-2')]
        k=0
        for i,j in itertools.product(range(subplot[0]), range(subplot[1])):

            if final_data[column_to_plot[k]].dtype == 'category':
                ax[i,j].scatter(x = final_data['umap-1'], y = final_data['umap-2'], s=s,
                                c=final_data[column_to_plot[k]].map(all_cat_colormap),
                                cmap=cmap, alpha=alpha, **kwargs)
                # create legend
                patchList = []
                for key in list(final_data[column_to_plot[k]].unique()):
                    data_key = mpatches.Patch(color=all_cat_colormap[key], label=key)
                    patchList.append(data_key)    
                    ax[i,j].legend(handles=patchList,bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)
            else:               
                im = ax[i,j].scatter(x = final_data['umap-1'], y = final_data['umap-2'], s=s,
                                c=final_data[column_to_plot[k]],
                                cmap=cmap, alpha=alpha, **kwargs)
                plt.colorbar(im, ax=ax[i, j])

            ax[i,j].tick_params(right= False,top= False,left= False, bottom= False)
            ax[i,j].set_xticklabels([]); ax[i,j].set_yticklabels([])
            ax[i,j].set_xlabel("UMAP-1"); ax[i,j].set_ylabel("UMAP-2")
            ax[i,j].set_title(column_to_plot[k])
            if tight_layout is True:
                plt.tight_layout()
            k = k+1 # iterator

    # if save figure is requested
    if save_figure is not None:
        plt.savefig(save_figure)  

    # return data if needed
    if return_data is True:
        return final_data