A set of CLI-based data filtering and plotting tools for proteomics data at BCM.
This package is written for Python3. Nothing should be prevented from getting it to work with Python2 as well, but it will likely not work without changes. Pull requests welcome.
- numpy
- scipy
- scikit-learn
- pandas
- matplotlib
- seaborn
- click
- bcmproteomics
- bcmproteomics_ext
Clone the repository and use pip to install:
git clone https://github.com/asalt/correlation-plotter
cd ./correlation-plotter
pip install .correlationplot is a suite of tools which do different things with the
data. Each tool is specified by the user at the command line after the base
correlationplot command (see below for examples).
The following tools are available:
- cluster
- Cluster the data via either hierarchical or K-means clustering and produce a heatmap
- export
- export the data after specified filtering
- pca
- Produce a PCA plot with the first two principle components from the data.
- scatter
- Produce a grid of scatterplots comparing each sample to others.
The first command has a required input of EXPERIMENT_FILE, a config file
that specifies which experiments to download and analyze. Additionally,
optional arguments can be set such as whether or not to perform iFOT
normalization and how many non-zero observations are to be permitted per
observation.
Usage: correlationplot [OPTIONS] EXPERIMENT_FILE COMMAND1 [ARGS]... [COMMAND2
[ARGS]...]...
Options:
--additional-info PATH .ini file with metadata for isobaric data used
for scatter and PCA plots
--data-dir DIRECTORY location to store and read e2g files [default:
./data/]
--funcats TEXT Optional gene subset based on funcat or funcats,
regular expression allowed.
--iFOT Calculate iFOT (divide by total input per
experiment) [default: False]
-n, --name TEXT An optional name for the analysis that will place
all results in a subfolder.
--non-zeros INTEGER Minimum number of non-non_zeros allowed across
samples. [default: 0]
--taxon [human|mouse|all] [default: all]
--help Show this message and exit.
Commands:
cluster
export
pca
scatter
The experiment file is a required input for specifying which experiments to analyze. It has the following format:
[NAME 1] recno = 1 runno = 1 searchno = 1
Note that runno and searchno will default to 1 if not specified. The name within the brackets will be used to label the sample in all plots and data exports.
Additionally, metadata can be added for each experiment. For example, one may wish to specify which samples are controls and which are treatments. The modified version of the config file will then look like this:
[NAME 1] recno = 1 runno = 1 searchno = 1 Treatment = Vehicle Batch = 1 [NAME 2] recno = 2 runno = 1 searchno = 1 Treatment = Drug Batch = 1
Here we have added Treatment and Batch labels for each entry. These will
be used as labels in the cluster subcommand.
Finally, the pca subcommand needs an additional specification for marking
the samples on the plot. Continuing with the above example, we can add this
to our config file:
[__PCA__] color = Treatment marker = Batch
Now the different treatments will be represented in different colors and different batches will be represented as different shapes in the PCA plot.
Note that the metadata can be specified with any desired name as long as it is consistently used for all experiments (case sensitive).
Usage: correlationplot [OPTIONS]
Options:
--col-cluster / --no-col-cluster
Cluster columns via hierarchical clustering.
Note this is overridden by specifying
`nclusters` [default: True]
--gene-symbols Show Gene Symbols on clustermap [default:
False]
--geneids PATH Optional list of geneids to subset by.
Should have 1 geneid per line.
--highlight-geneids PATH Optional list of geneids to highlight by.
Should have 1 geneid per line.
--nclusters TEXT If specified by an integer, use that number
of clusters via k-means clustering. If
specified as `auto`, will try to find the
optimal number of clusters
--row-cluster / --no-row-cluster
Cluster rows via hierarchical clustering
[default: True]
--seed TEXT seed for kmeans clustering
--show-metadata Show metadata on clustermap if present
[default: False]
--standard-scale [None|0|1] [default: None]
--z-score [None|0|1] [default: 0]
--help Show this message and exit.
After calling the main command, the cluster subcommand is specified.:
correlationplot ./example.config clusterBy default, this command clusters samples and gene products via hierarchical
clustering and produces the resulting heatmap. Alternatively, the command can
be clustered via K-means or DBSCAN if specified.
K-means clustering can be used by specifying --nclusters as the
number of desired clusters or auto. If auto is specified, the optimal
number of clusters will be estimated by using the number of clusters with the
highest silhouette score (see
https://en.wikipedia.org/wiki/Silhouette_(clustering) for more information).
A plot of average silhouette scores across the various number of clusters
used is produced when auto is specified. This plot should be examined to
check for the case where multiple numbers of clusters have similar silhouette scores.
KMeans clustering returns a heatmap with the gene products grouped by the
resulting clusters. It also exports the data annotated with the corresponding
clusters, as well as a silhouette analysis plot for each cluster.
DBSCAN similarly returns a heatmap with the gene products annotated by cluster. Black indicates data that does not fall into any of the observed clusters. It also exports the data annotated with the cluster information as well as a silhouette analysis plot.
Run this subcommand:
correlationplot ./example.config pca
Run this subcommand:
correlationplot ./example.config scatter
Run this subcommand:
correlationplot ./example.config export
Multiple subcommands can be chained together if multiple subcommands are to be run:
correlationplot --non-zeros 4 --name example ./example.config cluster --nclusters auto export --level area
Notice that the optional arguments are specified for each command and subcommand.