gibbsdavidl / genesignalproc

Application of graph signal processing to gene expression data.

GeneSignalProc

Application of graph signal processing to gene expression data. Single sample gene set scoring.

0. Build a gene-gene functional graph from a collection of gene sets. a.) a sparse matrix is returned representing the adjacency matrix b.) a list of genes is returned, which are the row labels for the adjacency matrix c.) also saves a python pickle of igraph objects and gene ID dictionary d.) can use a gene lists to control the 'gene vocabulary' of the graph

1. Filter the gene expression data. a.) uses a matrix of gene expression, samples in rows, genes in cols b.) and a graph constructed from gene sets (src/see makeGraphs.py) c.) data is filtered to match the gene labels of the graph

2. Score samples for each gene set. a.) Scores are internally normalized by comparison to a background of subsampled gene sets (from gene set graph).

3. Compare the gene set scores a.) based on predictive capability using RandomForests b.) rank the sets based on cross validation

======= To Run:

python3.6 GeneSignalProc/src/main.py --help

python3.6 GeneSignalProc/src/main.py

    print("makegraphs args: ")
    print("    -m Mode")
    print("    -d data dir")
    print("    -o output dir")
    print("    -g genesets file as .gmt")
    print("    -a adjacency file if available")
    print("    -e gene list file if available")
    print("    -n number of subgraphs")
    print("    -x max size of subgraphs")
    print("    -t threshold for geneset overlaps")
    print("    -c number of cores")

python3.6 GeneSignalProc/src/main.py

    print("setscoring args: ")
    print("    -m Mode")
    print("    -d data dir")
    print("    -o output dir")
    print("    -r expression data, samples in rows, genes in columns, first col is sample name, first row is gene names")
    print("    -p phenotype file if available, same order as expression data")
    print("    -a adjacency file")
    print("    -s subgraph file")
    print("    -t threshold for geneset overlaps")
    print("    -x threshold for graph edges")
    print("    -e gene list")
    print("    -g genesets file as .gmt")
    print("    -f filter name")
    print("    -l number of scale levels")
    print("    -c num cores")

Examples:

Working with the set of Cancer Hallmark gene sets.

python3.6 GeneSignalProc/src/main.py -m makegraphs -d /gene_set_work/ -g h.all.v6.2.symbols.gmt -c 4 -a h.all.v6.2.symbols.gmt_adjmat.tsv.gz -e h.all.v6.2.symbols.gmt_genes.tsv.gz -n 10 -x 7

python3.6 GeneSignalProc/src/main.py -m setscoring -d /gene_set_work/ -r ivy20t.tsv -g h.all.v6.2.symbols.gmt -a h.all.v6.2.symbols.gmt_adjmat.tsv.gz -s h.all.v6.2.symbols.gmt_subgraphs.tsv.gz -e h.all.v6.2.symbols.gmt_genes.tsv.gz -p ivy20_pheno.tsv -c 4 -l 10 -t 0.2 -x 0.05