BokehHeat

Abstract

Bokehheat provides a python3, bokeh based, interactive boolean data, categorical data, numerical data, dendrogram, and heatmap plotting implementation.

• Minimal requirement: python >= 3.6, bokeh >= 1.1
• Dependencies: bokeh, pandas, scipy, selenium, phantomjs, pillow
• Programmer: bue, jenny
• Date origin: 2018-08
• License: >= GPLv3
• User manual: this README file
• Source code: https://gitlab.com/biotransistor/bokehheat

Available bokehheat plots are:

• heat.cdendro: an interactive categorical dendrogram plot implementation.

• heat.bbar: an interactive boolean bar plot implementation.

• heat.cbar: an interactive categorical bar plot implementation.

• heat.qbar: an interactive quantitative bar plot implementation.

• heat.heatmap: an interactive heatmap implementation.

• heat.clustermap: an interactive cluster heatmap implementation which combines heat.cdendro, heat.bbar, heat.cbar, heat.qbar and heat.heatmap under the hood.

• jheat.jdendro: javatreeview compatible dendogram gtr, atr file output.

• jheat.jheatmap: javatreeview compatible heatmap cdt file output.

• jheat.jclustermap: javatreeview compatible heatmap cdt, gtr and atr file output, which runs jheat.jdendro and jheat.jheatmap under the hood.

Example Results

For the real interactive experience please clone or download this repository and open theclustermap_0.0.0.html file with your favorite web browser (we recommend FireFox) or install bokehheat and run this tutorial.

Figure: This is a poor, static heat.clustermap png version.

HowTo Guide

How to install bokehheat?

pip3 install bokehheat

How to load the bokehheat library?

from bokehheat import heat

How to get reference information about how to use each bokehheat module?

from bokehheat import heat

help(heat.cdendro)
help(heat.bbar)
help(heat.cbar)
help(heat.qbar)
help(heat.heatmap)
help(heat.clustermap)

How to get reference information about how to use each javatreeview compatible module?

from bokehheat import jheat

help(jheat.jdendro)
help(jheat.jheatmap)
help(jheat.jclustermap)

How to integrate bokehheat plots into pweave documents?

from pweave.bokeh import output_pweave, show

output_pweave()
o_clustermap, ls_xaxis, ls_yaxis = heat.clustermap(...)
show(o_clustermap)

How to integrate bokehheat plots into Jupyter Notebook and Lab?

Please, have a look at this page from the official bokeh documentaion.

Tutorial

This tutorial guides you through a cluster heatmap generation process.

1. Load libraries needed for this tutorial:

   # library
   from bokehheat import heat, jheat
   from bokeh.io import show
   from bokeh.palettes import Reds9, RdBu11, YlGn8, Colorblind8
   import numpy as np
   import pandas as pd
   

2. Prepare data:

   # generate test data
   ls_sample = ['sampleA','sampleB','sampleC','sampleD','sampleE','sampleF','sampleG','sampleH']
   ls_variable = ['geneA','geneB','geneC','geneD','geneE','geneF','geneG','geneH', 'geneI']
   ar_z = np.random.rand(9,8)
   df_matrix = pd.DataFrame(ar_z)
   df_matrix.index = ls_variable
   df_matrix.columns = ls_sample
   df_matrix.index.name = 'y'
   df_matrix.columns.name = 'x'
   
   # generate some gene annotation
   df_variable = pd.DataFrame({
       'y': ls_variable,
       'genereal': list(np.random.random(9) * 2 - 1),
       'genetype': ['Ligand','Ligand','Ligand','Ligand','Ligand','Ligand','Receptor','Receptor','Receptor'],
       'genetype_color': ['Cyan','Cyan','Cyan','Cyan','Cyan','Cyan','Cornflowerblue','Cornflowerblue','Cornflowerblue'],
       'geneboole': [False, False, False, True, True, True, False, False, False],
   })
   df_variable.index = df_variable.y
   
   # generate some sample annotation
   df_sample = pd.DataFrame({
       'x': ls_sample,
       'age_year': list(np.random.randint(0,101, 8)),
       'sampletype': ['LumA','LumA','LumA','LumB','LumB','Basal','Basal','Basal'],
       'sampletype_color': ['Purple','Purple','Purple','Magenta','Magenta','Orange','Orange','Orange'],
       'sampleboole': [False, False, True, True, True, True, False, False],
   })
   df_sample.index = df_sample.x
   

3. Generate categorical and quantitative sample and gene annotation tuple of tuples:

   t_yboole = (df_variable,['geneboole'],'Red','Maroon') # True, False
   t_ycat = (df_variable, ['genetype'], ['genetype_color'])
   t_yquant = (df_variable, ['genereal'], [-1], [1], [Colorblind8][::-1])
   t_xboole = (df_sample,['sampleboole'],'Red','Maroon') # True, False
   t_xcat = (df_sample, ['sampletype'], ['sampletype_color'])
   t_xquant = (df_sample, ['age_year'], [0], [128], [YlGn8][::-1])
   tt_boolecatquant = (t_yboole, t_ycat, t_yquant, t_xboole, t_xcat, t_xquant)
   

4. Generate the cluster heatmap:

   s_file = "theclustermap.html"  # or "theclustermap.png"
   o_clustermap, ls_xaxis, ls_yaxis = heat.clustermap(
       df_matrix = df_matrix,
       ls_color_palette = Reds9,
       r_low = 0,
       r_high = 1,
       s_z = "log2",
       tt_axis_annot = tt_boolecatquant,
       b_ydendo = True,
       b_xdendo = True,
       #s_method='average',
       #s_metric='euclidean',
       #b_optimal_ordering=True,
       #i_px = 64,
       #i_height = 12,
       #i_width = 12,
       #i_min_border_px = 128,
       s_filename=s_file,
       s_filetitel="the Clustermap",
   )
   

5. Display the result:

   print(f"check out: {s_file}")
   print(f"y axis is: {ls_yaxis}")
   print(f"x axis is: {ls_xaxis}")
   
   show(o_clustermap)
   

The resulting clustermap should look something like the example result in the section above.

1. Generate cdt, gtr, atr files to be able to study heatmap and clustering in the JavaTreeView and TreeView3 software.

   t_out = jheat.jclustermap(
       df_matrix=df_matrix,
       tt_axis_annot = tt_boolecatquant,
       s_xcolor = "age_year",
       s_ycolor = "genetype",
       b_xdendo = True,
       b_ydendo = True,
       #s_method = 'average',
       #s_metric = 'euclidean',
       #b_optimal_ordering = True,
       s_filename = "jclustermap",
   )
   print(t_out)
   

Discussion

In bioinformatics a clustered heatmap is a common plot to present gene expression data from many patient samples. There are well established open source clustering software kits like Cluster and TreeView, JavaTreeView, and TreeView3 for producing and investigating such heatmaps.

Static cluster heaptmap implementations

There exist a wealth of R and R/bioconductor packages with static cluster heatmaps functions (e.g. heatmap.2 from the gplots library), each one with his own pros and cons.

In Python the static cluster heatmap landscape looks much more deserted. There are some ancient mathplotlib based implementations like this active state recipe or the heatmapcluster library, or the hclustering library. There is the seaborn clustermap implementation, which looks good but might need hours of tweaking to get an agreeable plot with all the needed information out.

So, static heatmaps are not really a tool for exploring data.

Interactive cluster heatmap implementations

There exist d3heatmap a R/d3.js based interactive cluster heatmap packages. And heatmaply, a R/plotly based package. Or on a more basic level R/plotly based cluster heatmaps can be written with the ggdendro and ggplot2 library.

But I have not found a full fledged python based interactive cluster heatmap library. Neither Python/plottly nor Python/bokeh based. The only Python/bokeh based cluster heatmap implementation I was really aware of was this listing from Daniel Russo. Later on I found this bokeh based bkheatmap implementation from Wen-Wei Liao.

Synopsis

All in all, all of these implementations were not really what I was looking for. That is why I rolled my own. Bokehheat is a Python3/bokeh based interactive cluster heatmap library.

The challenges this implementation tried to solve are, the library should be:

• easy to use with pandas dataframes.
• static output, this means there have to be an easy way to generate static png files as output.
• interactive output, this means there have to be a easy way to generate hover and zoomable plots.
• output should be stored in computer platform independent and easy accessible format, like png files or java script spiced up html file, which can be opened in any webbrowser.
• possibility to add as many boolean, categorical, and quantitative y and x annotation bars as wished.
• possibility to hierarchical cluster y and/or x axis.
• snappy interactivity, even with big datasets with lot of samples and genes. (It turns out bokehheat is ok with hundreds of samples and genes but not with thousands. This is why the jheat.py extension was added, to be easily able to generate JavaTreeView and TreeView3 compatible output.)

Further directions

If you are interested in data visualization, check out Jake VanderPlas talk Python Visualization Landscape from the PyCon 2017 in Portland Oregon (USA).

Contributions

• Implementation: Elmar Bucher
• Documentation: Jennifer Eng, Elmar Bucher
• Helpful discussion: Mark Dane, Daniel Derrick, Hongmei Zhang, Annette Kolodize, Jim Korkola, Laura Heiser, Matt Melnicki, Bryan Van de Ven, and Daniele Procida.