clustree 0.2.1

Visualize relationship between clusterings at different resolutions

clustree

Status

Functionality: Implemented

• Directed graph representing clustree. Nodes are parsed images and node information is encoded by a border surrounding the image.
• Loading: Data provided directly or through a path to parent directory. Images provided through a path to parent directory.
• Appearance: Edge and node color can correspond to one of: #samples that pass through edge/node, cluster resolution K, or a fixed color. In the case of node color, a column name in the data and aggregate function can be used too. Use of column name and #samples creates a continuous colormap, whilst the other options result in discrete colors.
• Layout: Reingold-Tilford algorithm used for node positioning. Not recommended for kk > 12 due to memory bottleneck in igraph dependency.
• Legend: demonstration of node / edge color.

Functionality: To Add

• Legend: demonstration of transparency of edges.
• Layout: Bespoke implementation of Reingold-Tilford algorithm to overcome dependency's memory bottleneck.

Usage

Installation

Install the package with pip:

pip install clustree

Quickstart

The powerhouse function of the library is clustree. Use

from clustree import clustree

to import the function. A detailed description of the parameters is provided below.

def clustree(
    data: Union[Path, str],
    prefix: str,
    images: Union[Path, str],
    output_path: Optional[Union[Path, str]] = None,
    draw: bool = True,
    node_color: str = "prefix",
    node_color_aggr: Optional[Union[Callable, str]] = None,
    node_cmap: Union[mpl.colors.Colormap, str] = "inferno",
    edge_color: str = "samples",
    edge_cmap: Union[mpl.colors.Colormap, str] = "viridis",
    orientation: Literal["vertical", "horizontal"] = "vertical",
    layout_reingold_tilford: bool = None,
    min_cluster_number: Literal[0, 1] = 1,
    border_size: float = 0.05,
    figsize: tuple[float, float] = None,
    arrows: bool = None,
    node_size: float = 300,
    node_size_edge: Optional[float] = None,
    dpi: float = 500,
    kk: Optional[int] = None,
) -> DiGraph:
    """

• data : Path of csv or DataFrame object.
• prefix : String indicating columns containing clustering information.
• images : Path of directory that contains images.
• output_path : Absolute path to save clustree drawing at. If file extension is supplied, must be .png. If None, then output not written to file.
• draw : Whether to draw the clustree. Defaults to True. If False and output_path supplied, will be overridden.
• node_color : For continuous colormap, use 'samples' or the name of a metadata column to color nodes by. For discrete colors, use 'prefix' to color by resolution or specify a fixed color (see Specifying colors in Matplotlib tutorial here: https://matplotlib.org/stable/tutorials/colors/colors.html). If None, default set equal to value of prefix to color by resolution.
• node_color_aggr : If node_color is a column name then a function or string giving the name of a function to aggregate that column for samples in each cluster.
• node_cmap : If node_color is 'samples' or a column name then a colourmap to use (see Colormap Matplotlib tutorial here: https://matplotlib.org/stable/tutorials/colors/colormaps.html).
• edge_color : For continuous colormap, use 'samples'. For discrete colors, use 'prefix' to color by resolution or specify a fixed color (see Specifying colors in Matplotlib tutorial here: https://matplotlib.org/stable/tutorials/colors/colors.html). If None, default set to 'samples'.
• edge_cmap : If edge_color is 'samples' then a colourmap to use (see Colormap Matplotlib tutorial here: https://matplotlib.org/stable/tutorials/colors/colormaps.html).
• orientation : Orientation of clustree drawing. Defaults to 'vertical'.
• layout_reingold_tilford : Whether to use the Reingold-Tilford algorithm for node positioning. Defaults to True if (kk <= 12), False otherwise. Setting True not recommended if (kk > 12) due to memory bottleneck in igraph dependency.
• min_cluster_number : Cluster number can take values (0, ..., K-1) or (1, ..., K). If the former option is preferred, parameter should take value 0, and 1 otherwise. Defaults to None, in which case, minimum cluster number is found automatically.
• border_size : Border width as proportion of image width. Defaults to 0.05.
• figsize : Parsed to matplotlib to determine figure size. Defaults to (kk/2, kk/2), clipped to a minimum of (3,3) and maximum of (10,10).
• arrows : Whether to add arrows to graph edges. Removing arrows alleviates appearance issue caused by arrows overlapping nodes. Defaults to True.
• node_size : Size of nodes in clustree graph drawing. Parsed directly to networkx.draw_networkx_nodes. Default to 300.
• node_size_edge: Controls edge start and end point. Parsed directly to networkx.draw_networkx_edges.
• dpi : Controls resolution of output if saved to file.
• kk : Choose custom depth of clustree graph.

Glossary

• cluster resolution: Upper case K. For example, at cluster resolution K=2 data is clustered into 2 distinct clusters.
• cluster number: Lower case k. For example, at cluster resolution 2 data is clustered into 2 distinct clusters k=1 and k=2.
• kk: highest value of K (cluster resolution) shown in clustree.
• cluster membership: The association between data points and cluster numbers for fixed cluster resolution. For example, [1, 1, 2, 2, 2] would mean the first 2 data points belong to cluster number 1 and the following 3 data points belong to cluster number 2.