MatNet: Construction of correlation network from a matrix

In bingzhang16/NetSAM: Network Seriation And Modularization

Construction of correlation network from a matrix

Description

The MatNet function can use one of three different methods to construct correlation network based on the input data matrix. The output correlation network can be used as an input of NetSAM function to identify hierarchical modules.

Usage

MatNet(inputMat, collapse_mode="maxSD", naPer=0.7, meanPer=0.8, varPer=0.8, corrType="spearman", matNetMethod="rank", valueThr=0.5, rankBest=0.003, networkType="signed", netFDRMethod="BH", netFDRThr=0.05, idNumThr=(-1),nThreads=3)

Arguments

inputMat	inputMat should contain a file name with extension "cct" or "cbt" or a matrix or data.frame object in R. The first column and first row of the "cct" or "cbt" file should be the row and column names, respectively and other parts are the numeric values. The detail information of "cct" or "cbt" format can be found in the manual of NetGestalt (www.netgestalt.org). A matrix or data.frame object should have row and column names and only contain numeric or integer values.
collapse_mode	If the input matrix data contains the duplicate ids, the function will collapse duplicate ids based on the collapse_mode. "mean", "median", "maxSD" and "maxIQR" represent the mean, median, max standard deviation or max interquartile range of id values in each sample. The default is "maxSD".
naPer	To remove ids with missing values in most of samples, the function calculates the percentage of missing values in all samples for each id and removes ids with over naPer missing values in all samples. The default naPer is 0.7.
meanPer	To remove ids with low values, the function calculates the mean of values for each id in all samples and remains top meanPer ids based on the mean. The default meanPer is 0.8.
varPer	Based on the remained ids filtered by meanPer, the function can also remove less variable ids by calculating the standard deviation of values for each id in all samples and remaining top varPer ids based on the standard deviation. The default varPer is 0.8.
corrType	The method to calculate correlation coefficient for each pair of ids. The function supports "spearman" (default) or "pearson" method.
matNetMethod	MatNet function supports three methods to construct correlation network: "value", "rank" and "directed". 1. "value" method: the correlation network only remains id pairs with correlations over cutoff threshold valueThr; 2. "rank" method: for each id A, the function first selects ids that significantly correlate with id A and then extracts a set of candidate neighbors (the number of ids is calculated based on rankBest) from the significant set that are most similar to id A. Then, for each id B in the candidate neighbors of id A, the function also extracts the same number of ids that are significant correlated and most similar to id B. If id A is also the candidate neighbors of id B, there will be an edge between id A and id B. Combining all edges can construct a correlation network; 3. "directed" method: the function will only remain the best significant id for each id as the edge.Combining all edges can construct a directed correlation network.
valueThr	Correlation cutoff threshold for "value" method. The default is 0.5.
rankBest	The percentage of ids that are most similar to one id for "rank" method. The default is 0.003 which means the "rank" method will select top 30 most similar ids for each id if the number of ids in the matrix is 10,000.
networkType	If networkType is "unsigned", the correlation of all pairs of ids will be changed to absolute values. The default is "signed".
netFDRMethod	p value adjustment methods for "rank" and "directed" methods. The default is "BH".
netFDRThr	fdr threshold for identifying significant pairs for "rank" and "directed" methods. The default is 0.05
idNumThr	If the matrix contains too many ids, it will take a long time and use a lot of memory to identify the modules. Thus, the function provides the option to set the threshold of number of ids for further analysis. After filtering by meanPer and varPer, if the number of ids is still larger than idNumThr, the function will select top idNumThr ids with the largest variance. The default is -1, which means there is no limitation for the matrix.
nThreads	MatNet function supports parallel computing based on multiple cores. The default is 3.

Note

For data with missing values, the function will take longer time to calculate correlation between each pair of ids than data without missing value.

Author(s)

Jing Wang

Examples

	inputMatDir <- system.file("extdata","exampleExpressionData.cct",package="NetSAM")
	matNetwork <- MatNet(inputMat=inputMatDir, collapse_mode="maxSD", naPer=0.7, meanPer=0.8, varPer=0.8, corrType="spearman", matNetMethod="rank", valueThr=0.6, rankBest=0.003, networkType="signed", netFDRMethod="BH", netFDRThr=0.05, idNumThr=(-1),nThreads=3)

bingzhang16/NetSAM documentation built on April 3, 2024, 3:35 a.m.