Grinn

fetchGrinnPtCorrNetwork - combine a grinn network queried from Grinn internal database to a partial correlation network

Description

from the list of keywords and input omics data e.g. normalized expression data or metabolomics data, it is a one step function to:
1. Build an integrated network (grinn network) by connecting these keywords to a specified node type, see fetchGrinnNetwork. The keywords can be any of these node types: metabolite, protein, gene and pathway. The Grinn internal database contains the networks of the following types that can be quried: metabolite-protein, metabolite-protein-gene, metabolite-pathway, protein-gene, protein-pathway and gene-pathway.
2. Compute a partial correlation network of input omics data, see datX. Correlation coefficients, pvalues and relation directions are calculated using qpgraph functions. The correlation coefficients are continuous values between -1 (negative correlation) and 1 (positive correlation), with numbers close to 1 or -1, meaning very closely correlated.
3. Combine the grinn network to the correlation network.

Usage

fetchGrinnCorrNetwork(txtInput, from, to, filterSource, returnAs, dbXref, datX, corrCoef, pval, alpha, epsilon, matrix.completion)

Arguments

txtInput	list of keywords containing keyword ids e.g. txtInput = list('id1', 'id2'). The keyword ids are from the specified database, see dbXref. Default is grinn id e.g. G371.
from	string of start node. It can be one of "metabolite","protein","gene","pathway".
to	string of end node. It can be one of "metabolite","protein","gene","pathway".
filterSource	string or list of pathway databases. The argument is required, if from or to = "pathway", see from and to. The argument value can be any of "SMPDB","KEGG","REACTOME" or combination of them e.g. list("KEGG","REACTOME").
returnAs	string of output type. Specify the type of the returned network. It can be one of "tab","json","cytoscape", default is "tab". "cytoscape" is the format used in Cytoscape.js.
dbXref	string of database name. Specify the database name used for the txtInput ids, see txtInput. It can be one of "grinn","chebi","kegg","pubchem","inchi","hmdb","smpdb","reactome","uniprot","ensembl","entrezgene". Default is "grinn". If pubchem is used, it has to be pubchem SID (substance ID).
datX	data frame containing normalized, quantified omics data e.g. expression data, metabolite intensities. Columns correspond to entities e.g. genes, metabolites, and rows to samples e.g. normals, tumors. Require 'nodetype' at the first row to indicate the type of entities in each column. See below for details.
corrCoef	numerical value to define the minimum value of absolute correlation, from 0 to 1, to include edges in the output.
pval	numerical value to define the maximum value of pvalues, to include edges in the output.
alpha	a numeric value specifying significance level of each test used in qpAvgNrr.
epsilon	a numeric value specifying the maximum cutoff value of the non-rejection rate met by the edges that are included in the qp-graph, see qpGraph.
matrix.completion	a string specifying algorithm to employ in the matrix completion operations used in qpPAC.

Details

datX is matrix in which rows are samples and columns are entities. The correlations of the columns of datX are computed.

Value

list of nodes and edges. The list is with the following componens: edges and nodes. Return empty list if found nothing.

Examples


# Create metabolite-gene network from the list of metabolites using grinn ids and combine the grinn network to a correlation network of metabolites
kw <- c('G160','G300','G371')
dummy <- rbind(nodetype=rep("metabolite"),t(mtcars))
colnames(dummy) <- c('G1.1','G27967','G371','G4.1',paste0('G',sample(400:22000, 28)))
result <- fetchGrinnPtCorrNetwork(txtInput=kw, from="metabolite", to="gene", datX=dummy, corrCoef=0.7, pval=0.05)
library(igraph)
plot(graph.data.frame(result$edges[,1:2], directed=FALSE))

References

Partial correlation-based analyses apply methods from the following publications:

Castelo R, et al. A robust procedure for Gaussian graphical model search from microarray data with p larger than n. Mach. Learn. Res., 7:2621-2650.
Castelo R, et al. Reverse engineering molecular regulatory networks from microarray data with qp-graphs. J Comput Biol, 16(2), pp. 213-27.