GSEA-based Programs for NGS data

This repository contains Python programs and example input file(s) from publicly available NGS datasets acquired from Gene Expression Omnibus to reproduce the Gene Set Enrichment Analysis (GSEA)-based approach for gene identification, verification, and comparison presented in the Park & Harris papers (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8342995/ and https://www.cscjournals.org/library/manuscriptinfo.php?mc=IJBB-262).

Gene Identification:

GSEA1datasetGeneID.py inputs one dataset containing clean and complete NGS data in tab-delimited .txt file format

GSEA2datasetsGeneID.py inputs two independent datasets containing clean and complete NGS data both in tab-delimited .txt file format

Both GSEAGeneID programs perform the following:

1) Z-score normalization of data across all samples for each gene in a dataset

2) Generation of gene signatures via T-score - returns two gene lists, one per dataset, with associated T-scores and p-values

3) Generation of query gene sets - returns a .gmt file (https://software.broadinstitute.org/cancer/software/gsea/wiki/index.php/Data_formats) that contains 4 query sets representing the most over- and under-expressed genes from each gene signature (default is 500 genes per query set)

4) GSEA between gene signatures and query gene sets - returns enrichment data (normalized scores, nominal p-values, plots) and leading-edge gene identification

NOTE: These algorithms compute T-score per scipy.stats.ttest_ind (https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.ttest_ind.html). This algorithm does not adjust T-score values using a minimum value for σ of .2 * absolute(μ), where μ=0 is adjusted to μ=1, like performed in the GSEA desktop version (https://www.gsea-msigdb.org/gsea/doc/GSEAUserGuideFrame.html).

Gene Verification/Comparison:

GSEACompare.zip contains 2 Python scripts inputing your identified gene panels and one dataset containing clean and complete NGS data in tab-delimited .txt file format

GSEAGeneVerify performs the following:

1) Z-score normalization of data across all samples for each gene in the dataset

2) Generation of gene signature via T-score - returns gene list with associated T-scores and p-values

3) GSEA between gene signature and identified gene panels - returns enrichment data (normalized scores, nominal p-values, plots) and leading-edge gene identification

4) Generation of random query gene sets - returns a .gmt file (https://software.broadinstitute.org/cancer/software/gsea/wiki/index.php/Data_formats) that contains specified number (1000 default) of query sets representing randomly selected genes from the gene signature

GSEAGeneRandom performs the following:

1) GSEA between gene signature and randomly generated gene panels - returns enrichment data (normalized enrichment scores only)

2) Box and whiskars plot of randomly generated normalized enrichment scores compared to achieved scores from identified gene panels

Required dependencies: numpy, pandas, scipy, and gseapy (https://gseapy.readthedocs.io/)

Limitations of these programs include:

1) Manual preparation of input files prior to program use to ensure clean and complete data that consistently uses the same gene identifier (e.g., probes, gene symbols, etc.) across all datasets

2) No GUI = manual manipulation of file names, column indicies, and query set sizes in the code as designated by the "#change ... as needed" comments

Please note that Python uses 0 for the index of the first column in a dataframe. Adjust your column counts accordingly.

3) Manual analysis of identified leading-edge genes and preparation of gene panels file, similar to .gmt format, for verification and comparison stages

4) Manual analysis to verify individual panel genes from identified leading-edge genes