clusterlinc-methods: Cluster Queried ncRNAs Based On Their Interaction Partners
In ManuelGoepferich/LINC: co-expression of lincRNAs and protein-coding genes

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

The function clusterlinc will give an overview of ncRNAs in a dataset. An input LINCmatrix will be converted to a LINCcluster. The following steps are conducted (I) computation of a correlation test, (II) setup of a distance matrix, (III) calculation of a dendrogram and (IV) selection of co-expressed genes for each query. The result is a cluster of ncRNAs and their associated protein-coding genes.

clusterlinc(linc,
            distMethod    = "dicedist",
            clustMethod   = "average",
            pvalCutOff    = 0.05,
            padjust       = "none",
            coExprCut     = NULL,
            cddCutOff     = 0.05,
            verbose       = TRUE)

`linc`	an object of the class `LINCmatrix`
`distMethod`	a method to compute the distance between ncRNAs; has to be one of `c("correlation", "pvalue", "dicedist")`
`clustMethod`	an algorithm to compute the dendrogram, has to be one of `c("ward.D", "ward.D2", "single", "complete", "average", "mcquitty", "median", "centroid")`
`pvalCutOff`	a threshold for the selection of co-expressed genes. Only protein-coding genes showing a significance in the correlation test lower than `pvalCutOff` will assigned to queried ncRNAs as interaction partner.
`padjust`	one of `c("holm", "hochberg", "hommel", "bonferroni", "BH", "BY", "fdr", "none")`
`coExprCut`	a single `integer` indicating the number of co-expressed genes to select. If this argument is used for each ncRNA the `coExprCut = n` protein-coding genes with the lowest p-value in the correlation test will be assigned to queries, respectively.
`cddCutOff`	a threshold that is only relevant for `distMethod = "dicedist"`. In this method `cddCutOff` defines whether a ncRNA and a protein-coding gene can be considered as interaction partners. This influences the distance matrix and the clustering process.
`verbose`	whether to give messages about the progression of the function `TRUE` or not `FALSE`

As a first step clusterlinc conducts the correlation test (stats::cor.test) using the correlation method and handeling of missing values inherited from the input LINCmatrix. Resulting p-values indicate the statistical robustness of correlations instead of absolute correlation values. Co-expression of ncRNAs to protein-coding genes is assumed if the p-value from the cor.test is lower than the given pvalCutOff. An alternative way to select co-expressed genes is provided by coExprCut. This argument has priority over pvalCutOff and can be used to pick the n genes with the lowest p-value for each ncRNA. In contrast to pvalCutOff, this will result in an equal number of assigned co-expressed genes. The argument padjust can be used for multiple testing correction. In most cases this is not compatible with distMethod = "dicedist".

For the computation of the distance matrix of ncRNA genes three methods can be applied. The first method "correlation" uses 1 - correlation as distance measure. In contrast, "pvalue" considers not the absolute correlation values, but p-values from the correlation test. A third method is termed "dicedist" and takes the Czekanovski dice distance [1] as distance measure. Here, the number of shared interaction partners between ncRNAs determines their relation to each other. The argument cddCutOff is an option to decide which p-values in the correlation matrix can be considered as interaction. A low threshold, for instance, will consider only interactions of ncRNAs and protein-coding genes supported by a p-value lower than the supplied threshold and therfore a robust correlation of these two genes. Based on the distance matrix a cluster of the ncRNAs will be computed by stats::hclust. Argument clustMethod defines which clustering method should be applied.

A LINCcluster can be recalculated with the command clusterlinc(LINCcluster, ...)) in order to change further arguments. plotlinc(LINCcluster, ...)) will plot a figure that shows the cluster of ncRNAs (dendrogram) and the number of co-expressed genes with respect to different thresholds. getbio(LINCcluster, ...)) will derive the biological terms associated with the co-expressed genes. Due to the correlation test longer calculation times can occur. A faster alternative to this function is singlelinc(). User-defined correlation test functions are supported for singlelinc() but not for clusterlinc().

an object of the class 'LINCmatrix' (S4) with 6 Slots

`results`	a `list` containing an object of the class `"phylo"` with the additional entry `neighbours`, a list of queries and co-expressed genes
`assignment`	a `character` vector of protein-coding genes
`correlation`	a `list` of `cormatrix`, the correlation of non-coding to protein-coding genes, `lnctolnc`, the correlation of non-coding to non-coding genes and `cortest`, p-values of the correlation test of non-coding to protein-coding genes
`expression`	the original expression matrix
`history`	a storage environment of important methods, objects and parameters used to create the object
`linCenvir`	a storage environment ensuring the compatibility to other objects of the `LINC` class

signature(linc = "LINCcluster"): (see details)
signature(linc = "LINCmatrix"): (see details)

plotlinc(LINCcluster, ...), getbio(LINCcluster, ...), ...

Manuel Goepferich

[1] Christine Brun, Francois Chevenet, David Martin, Jerome Wojcik, Alain Guenoche and Bernard Jacq" Functional classification of proteins for the prediction of cellular function from a protein-protein interaction network" (2003) Genome Biology, 5:R6.

linc ; singlelinc

data(BRAIN_EXPR)
class(crbl_matrix)

# call 'clusterlinc' with no further arguments
crbl_cluster <- clusterlinc(crbl_matrix)

# apply the distance method "correlation instead of "dicedist"
crbl_cluster_cor <- clusterlinc(crbl_matrix, distMethod = "correlation" )
# do the same as recursive call using the 'LINCcluster' object
# crbl_cluster_cor <- clusterlinc(crbl_cluster, distMethod = "correlation")

# select 25 genes with lowest p-values for each query
crbl_cluster_25 <- clusterlinc(crbl_matrix, coExprCut = 25)

# select onyl those with a p-value < 5e-5
crbl_cluster_5e5 <- clusterlinc(crbl_matrix, pvalCutOff = 5e-5)

# adjust for multiple testing
crbl_cluster_hochberg <- clusterlinc(crbl_matrix, distMethod = "correlation",
                                   padjust = "hochberg", pvalCutOff = 0.05)

# comparing two distance methods
 plotlinc(crbl_cluster)
 plotlinc(crbl_cluster_cor)