grass: grass
In lschen-stat/SNPath: Algorithms for pathway analysis using SNP data

Description Usage Arguments Details Value Author(s) References See Also Examples

This function calculates the p-value for the association between a pathway and disease risk using the GRASS algorithm. GRASS summarizes the genetic structure by singular value decomposition for each gene as eigenSNPs and uses a novel form of regularized regression technique, termed group ridge regression, to select representative eigenSNPs for each gene and assess their joint association with disease risk.

1 2	grass(cl=NULL, snp.dat, snp.info, gene.info, gene.set, y, weights=NULL, gene.def=c("abs", "rel"), dist=20000, k=1, B=100, nominal.p=FALSE)

`cl`	Cluster object for parallel computing. We recommand using the ‘snow’ and ‘Rmpi’ libraries to generate the clusters. See examples for illustration. If clusters are not provided, cl=NULL, computing will be conducted on the local machine. Parallel computing is optional but highly recommended.
`snp.dat`	A matrix with snp data. Data are coded as 0, 1, 2, corresponding to homozygotes for the major allele, heterozygotes, and homozygotes for the minor allele, respectively. Each row of the matrix is one SNP and each column represents one subject.
`snp.info`	A matrix with three columns: snp.Name (name of SNP), chr (chromosome), and pos (position where the SNP is located). It is suggested that the SNPs are sorted by their locations in the genome. Please insure that the same reference is used for SNP position and for gene position.
`gene.info`	A matrix with four columns with the order of gene.Name, chr, Start, and End. The name of the gene is under gene.Name. The chromosome is chr, and the start and end coordinate (start < end) are in the Start and End columns. It is suggested that the genes are sorted by their locations in the genome.
`gene.set`	A list of pathways. Each element of the list consists of the gene names of a pathway of interest.
`y`	A vector of case/control status. Cases are coded as 1 and controls are coded as 0.
`weights`	An optional numerical vector specifying each subject’s sample weight, the inverse probability that the subject is selected.
`gene.def`	An optional character string giving a method for assigning SNPs to the nearest genes by genomic distance. It can be "abs", which stands for absolute genomic distance in terms of bp, or "rel" for relative distance.
`dist`	A numerical value of absolute genomic distance (in base pair). Only used when gene.def = "abs". For example, gene.def="abs" and dist=20000, SNPs that are within 20kb of either end of a gene will be assigned to that gene.
`k`	A numerical value of relative genomic distance. Only used when gene.def = "rel". Each SNP is assigned to the k nearest genes on either side plus the gene that contains the SNP. A SNP can be assigned to a maximum of 2k+1* genes.
`B`	The number of permutations performed to calculate null statistics.
`nominal.p`	A logical value indicating if p-values should be calculated based on normal approximation of null statistics. The default is FALSE, which means p-values are calculated by permutation. When computation load is heavy, one can use a small number of permutations B and set nominal.p = TRUE.

The GRASS algorithm first summarizes the genetic structure (e.g., LD) for each gene as independent eigenSNPs via principal component analysis. Next, GRASS uses regularized regression technique with a novel form of penalty function, group ridge penalty, to estimate the association of eigenSNPs with disease risk. The group ridge regularized regression applies a lasso penalty within a gene to select representative eigenSNPs while simultanously using a ridge penalty among the genes to shrink the gene estimate. In this way, efficient estimation is achieved even with high-dimensional data and the pathway statistics are not dominated by a single gene or a single SNP. A pathway statistic is calculated from the penalized log odds ratios of eigenSNPs in the pathway. By permuting disease status, recomputing the null pathway statistics and comparing them with the observed ones, we obtain the p-value of disease association for each pathway.

Returns a numerical vector of p-values, corresponding to the pathways in gene.set.

Lin S. Chen <lche2@fhcrc.org>

Chen LS, Hutter CM, Potter JD, Liu Y, Prentice RL, Peters U, Hsu L (2010). Insights into colon cancer etiology using a regularized approach to gene set analysis of GWAS data. American Journal of Human Genetics, 86(6): 860-871.

aligator , gseaSnp , plinkSet

  library(SNPath)
  data(simDat)
  
  library(snow)
  cl <- makeCluster(c("localhost","localhost"), type = "SOCK")
  clusterEvalQ(cl, {
    library(SNPath)
  })

  path.pval <- grass(cl=cl, snp.dat=snpDat, snp.info=snp.info, gene.info=gene.info,
                   gene.set=sim.pathway, y=y, gene.def="rel", B=1000)
  
  stopCluster(cl)