blitzgsea 1.3.40

Package for fast and accurate calculation of Gene Set Enrichment Analysis (GSEA) similar to prerank using gamma distribution approximation.

Installation | Example | Optional Parameters | Speed-up | Plotting | Attribution | References

blitzGSEA Introduction

This Python package provides a computationally performant Gene Set Enrichment Analysis (GSEA) implementation of the pre-rank algorithm [1]. GSEApy was used as the reference for the running sum and enrichment score calculation [2]. The algorithm estimates the enrichment score (ES) distribution of the null model by fitting data to gamma distibutions instead of calculating permutations for each gene set. blitzGSEA calculates p-values with much higher accuracy than other reference implementations available in Python.

Gene set libraries can directly be loaded from Enrichr (https://maayanlab.cloud/Enrichr). For this use the blitzgsea.enrichr.get_library() function. All libraries can also be listed with blitzgsea.enrichr.print_libraries().

blitzGSEA provides plotting functions to generate publication ready figures similar to the original GSEA-P software. blitzgsea.plot.running_sum() plots an enrichment plot for a single gene set and blitzgsea.plot.top_table() plots the top n gene sets in a compact table.

Installation

blitzGSEA is currently only available as a Python package in this GitHub repository. You can install the blitzGSEA Python package and its dependencies through pip by using the following command:

$ pip install blitzgsea

Run enrichment analysis using blitzGSEA

blitzGSEA depends on two input files. 1) a gene signature and 2) a gene set library. The gene set library is a dictionary with the name of the gene set as key and a list of gene ids as values. Gene set libraries can be loaded directly from Enrichr. The signature should be a pandas dataframe with two columns [0,1]. The first column should contain the gene ids (matching the gene ids in the gene set library).

Python example

This short example will download two files (signature and gene set library). The gene set library consists of KEGG pathways and the signature is an example signature of differential gene expression of muscle samples from young and old donors. Differential gene expression was computed with Limma Voom.

import blitzgsea as blitz
import pandas as pd

# read signature as pandas dataframe
signature = pd.read_csv("https://github.com/MaayanLab/blitzgsea/raw/main/testing/ageing_muscle_gtex.tsv")

# list available gene set libraries in Enrichr
blitz.enrichr.print_libraries()

# use enrichr submodule to retrieve gene set library
library = blitz.enrichr.get_library("KEGG_2021_Human")

# run enrichment analysis
result = blitz.gsea(signature, library)

Example Input

index	0	1
1	ADO	-7.833439
2	CHUK	-7.800920
3	GOLGA4	-7.78722
...	...	...

The gene set library is a dictionary with the gene set names as key and lists of gene ids as values.

{
'ERBB2 SIGNALING PATHWAY (GO:0038128)': ['CDC37',
                                          'PTPN12',
                                          'PIK3CA',
                                          'SOS1',
                                          'CPNE3',
                                          'EGF',
                                          ...
                                         ],
'HETEROTYPIC CELL-CELL ADHESION (GO:0034113)': ['DSC2',
                                                 'ITGAV',
                                                 'ITGAD',
                                                 'LILRB2',
                                                 ...
                                                ],
...
}

Optional Parameters

The main function of blitzgsea.gsea() supports several optional parameters. The default parameters should work well for most use cases.

parameter name	type	default	description
permutations	int	2000	Number of randomized permutations to estimate ES distributions.
min_size	int	5	Minimum number of genes in geneset.
max_size	int	4000	Maximal number of genes in gene set.
anchors	int	20	Number of gene set size distributions calculated. Remaining are interpolated.
processes	int	4	Number of parallel threads. Not much gain after 4 threads.
symmetric	bool	False	Use same distribution parameters for negative and positive ES. If False estimate them separately.
signature_cache	bool	True	Cache precomputed anchor parameters in memory for later reuse.
shared_null	bool	False	Use same null for signatures if a compatible model already exists. (uses KL-divergence test).
kl_threshold	float	0.3	Controls how similar signature value distributions have to be for reuse.
kl_bins	int	200	Number of bins in PDF representation of distributions for KL test.
plotting	bool	False	Plot estimated anchor parametes.
verbose	bool	False	Toggle additonal output.
progress	bool	False	Toggle progress bar.
seed	int	0	Random seed. Same seed will result in identical result. If seed equal -1 generate random seed.
add_noise	bool	False	Add small random noise to signature. The noise is a fraction of the expression values.
center	bool	True	Center signature values at 0 before calculating running sum.

Speeding up enrichment calculations

blitzGSEA is currently the fastest GSEA implementation. The most time-consuming step of blitzGSEA is the generation of a robust null distribution to compute p-values. Since the null distribution depends on the value distribution of the input signature, blitzGSEA will, by default, compute a new null for each new input signature. blitzGSEA can compute the similarity between input signatures using Kullback–Leibler divergence to identify similar signatures to share null models. A cached null model is used if a previous signature has a similar value distribution. The relevant parameters of the blitzgsea.gsea() function are shown below:

parameter name	type	default	description
signature_cache	bool	True	Cache precomputed anchor parameters in memory for later reuse.
shared_null	bool	False	Use same null for signatures if a compatible model already exists. (uses KL-divergence test).
kl_threshold	float	0.3	Controls how similar signature value distributions have to be for reuse. The smaller the more conservative.
kl_bins	int	200	Number of bins in PDF representation of distributions for KL test.

Example

import blitzgsea as blitz
import pandas as pd

# read signature as pandas dataframe
signature = pd.read_csv("https://github.com/MaayanLab/blitzgsea/raw/main/testing/ageing_muscle_gtex.tsv")

# run enrichment analysis
result = blitz.gsea(signature, library, shared_null=True)

Plotting enrichment results

blitzGSEA supports several plotting functions. blitzgsea.plot.running_sum() and blitzgsea.plot.top_table() can be used after enrichment has been performed. blitzgsea.plot.running_sum() shows the running sum of an individual gene set. It has a compact mode in which the image will be more readable if small. blitzgsea.plot.top_table() shows the top n enriched gene sets and displays the results in a table, with normalized enrichment score (NES) and the distribution of hits relative to the gene ranking of the signature.

Example

import blitzgsea as blitz
import pandas as pd

# read signature as pandas dataframe
signature = pd.read_csv("https://github.com/MaayanLab/blitzgsea/raw/main/testing/ageing_muscle_gtex.tsv")

# use enrichr submodule to retrieve gene set library
library = blitz.enrichr.get_library("KEGG_2021_Human")

# run enrichment analysis
result = blitz.gsea(signature, library)

# plot the enrichment results and save to pdf
fig = blitz.plot.running_sum(signature, "Cell adhesion molecules", library, result=result, compact=False)
fig.savefig("running_sum.png", bbox_inches='tight')

fig_compact = blitz.plot.running_sum(signature, "Cell adhesion molecules", library, result=result, compact=True)
fig_compact.savefig("running_sum_compact.png", bbox_inches='tight')

fig_table = blitz.plot.top_table(signature, library, result, n=15)
fig_table.savefig("top_table.png", bbox_inches='tight')

# disable switching of interactive plotting, which can cause issues in jupyter notebooks
# will plot the figure into the notebook
fig = blitz.plot.running_sum(signature, "Cell adhesion molecules", library, result=result, compact=False, interactive_plot=True)
fig_table = blitz.plot.top_table(signature, library, result, n=15, interactive_plot=True)

The resulting plots will look like the examples below:

running_sum.pdf

running_sum_compact.pdf

top_table.pdf

Sample shuffling

This is the sample shuffling algorithm from GSEApy. It performs a t-test to build signatures for phenotype shuffled groups. The input is a gene expression dataframe, which should be normalized for library size. groups is a list containing 0 or 1 describing the corresponding group for the samples in exprs. The index of exprs are the gene ids matching the gene set library.

blitz.shuffle.gsea(exprs, library, groups, permutations=50, seed=1)

parameter name	type	default	description
exprs	pd.DataFrame	NA	Normalized gene expression matrix.
library	dictionary	NA	Gene set library.
groups	list	NA	Phenotype group labels of samples. Labels are 0 or 1.
permutations	int	1000	Number of permutations.
seed	int	1	Random state seed.

Dependencies

Python 3.6+

Attribution

The statistical tool was developed by the Ma'ayan Laboratory. When using blitzgsea please cite the following reference:

Lachmann, Alexander, Zhuorui Xie, and Avi Ma’ayan. "blitzGSEA: efficient computation of gene set enrichment analysis through gamma distribution approximation." Bioinformatics (2022). https://academic.oup.com/bioinformatics/advance-article/doi/10.1093/bioinformatics/btac076/6526383?login=false

References

[1] Lachmann, Alexander, Zhuorui Xie, and Avi Ma’ayan. "blitzGSEA: efficient computation of gene set enrichment analysis through gamma distribution approximation." Bioinformatics (2022).

[2] Subramanian, Aravind, Heidi Kuehn, Joshua Gould, Pablo Tamayo, and Jill P. Mesirov. "GSEA-P: a desktop application for Gene Set Enrichment Analysis." Bioinformatics 23, no. 23 (2007): 3251-3253.

[3] Fang, Zhuoqing, Xinyuan Liu, and Gary Peltz. "GSEApy: a comprehensive package for performing gene set enrichment analysis in Python." Bioinformatics (2022).