adjclust
Adjacency-Constrained Clustering of a Block-Diagonal Similarity Matrix
Implements a constrained version of hierarchical agglomerative clustering, in which each observation is associated to a position, and only adjacent clusters can be merged. Typical application fields in bioinformatics include Genome-Wide Association Studies or Hi-C data analysis, where the similarity between items is a decreasing function of their genomic distance. Taking advantage of this feature, the implemented algorithm is time and memory efficient. This algorithm is described in Chapter 4 of Alia Dehman (2015) < https://hal.archives-ouvertes.fr/tel-01288568v1>.
adjclust is a package that provides methods to perform adjacency-constrained hierarchical agglomerative clustering. Adjacency-constrained hierarchical agglomerative clustering is hierarchical agglomerative clustering (HAC) in which each observation is associated to a position, and the clustering is constrained so as only adjacent clusters are merged. It is useful in bioinformatics (e.g. Genome Wide Association Studies or Hi-C data analysis).
adjclust provides three user level functions: adjClust, snpClust and hicClust, which are briefly explained below.
Installation
You can install adjclust from github with:
devtools::install_github("pneuvial/adjclust")
adjClust
adjClust performs adjacency-constrained HAC for standard and sparse, similarity and dissimilarity matrices and dist objects. Matrix::dgCMatrix and Matrix::dsCMatrix are the supported sparse matrix classes. Let's look at a basic example
library("adjclust")sim <- matrix(c(1.0, 0.5, 0.2, 0.1,0.5, 1.0, 0.1, 0.2,0.2, 0.1, 1.0, 0.6,0.1, 0.2 ,0.6 ,1.0), nrow=4)h <- 3fit <- adjClust(sim, "similarity", h)plot(fit)
The result is of class chac. It can be plotted as a dendogram (as shown above). Successive merge and heights of clustering can be obtained by fit$merge and fit$height respectively.
snpClust
snpClust performs adjacency-constrained HAC for specific application of Genome Wide Association Studies (GWAS). A minimal example is given below. See GWAS Vignette for details.
library("snpStats")#> Loading required package: survival#> Loading required package: Matrixdata("ld.example", package = "snpStats")geno <- ceph.1mb[, -316] ## drop one SNP leading to one missing LD valueh <- 100ld.ceph <- ld(geno, stats = "R.squared", depth = h)image(ld.ceph, lwd = 0)
fit <- snpClust(geno, stats = "R.squared", h = h)#> Note: 132 merges with non increasing heights.plot(fit)#> Warning in plot.chac(fit):#> Detected reversals in dendrogram: mode = 'corrected', 'within-disp' or 'total-disp' might be more relevant.
sel_clust <- select(fit, "bs")plotSim(as.matrix(ld.ceph), clustering = sel_clust, dendro = fit)
hicClust
hicClust performs adjacency-constrained HAC for specific application of Hi-C data analysis. A minimal example is given below. See Hi-C Vignette for details.
library("HiTC")#> Warning: multiple methods tables found for 'acbind'#> Warning: multiple methods tables found for 'arbind'#> Warning: multiple methods tables found for 'rglist'
load(system.file("extdata", "hic_imr90_40_XX.rda", package = "adjclust"))binned <- binningC(hic_imr90_40_XX, binsize = 5e5)#> Bin size 'xgi' =500488 [1x500488]#> Bin size 'ygi' =500488 [1x500488]mapC(binned)#> minrange= 104 - maxrange= 36776.8
fitB <- hicClust(binned)#> Note: 5 merges with non increasing heights.plot(fitB)#> Warning in plot.chac(fitB):#> Detected reversals in dendrogram: mode = 'corrected', 'within-disp' or 'total-disp' might be more relevant.
plotSim(intdata(binned), dendro = fitB) # default: log scale for colors
Credits
Version 0.4.0 of this package was completed by Shubham Chaturvedi as a part of the Google Summer of Code 2017 program.
News
Version 0.5.7 [2018-09-26]
- Example Hi-C data now 10x smaller (subset of the original one). The package is smaller and tests are faster.
- implemented a model selection approach based on slope heuristic or on the broken stick heuristic to select a relevant number of clusters
- fixed minor problems in some method definition for class 'chac'
- proposed a log-transformation of data in the wrapper 'hicClust'
- implemented a heatmap with possible highlighting of the constrained clustering
- implemented an option to display number of the merge on the dendrogram
Version 0.5.6 [2018-02-08]
- changed dependencies to bioconductor packages 'HiTC' and 'snpStats' into Suggest and conditionnaly used them
Version 0.5.5 [2018-01-30]
- simplified code (replaced many C functions by a unique R function using Matrix)
- adjClust now properly handles similarities with diagonal entries different from 1
- removed arguments that were not used (blMin and verbose)
- simplified Hi-C example
Version 0.5.4 [2018-01-12]
- More tests for modify and modifySparse
- BUG FIX in condnCheck
Version 0.5.3 [2017-12-04]
- 'height' is now defined as the value of the linkage criterion (as is done in 'hclust'), rather than the total inertia of the clustering (as is done in 'rioja').
- Added several representations for the dendrogram corresponding to different choices for the height.
- Improved documentation and vignettes.
- Removed non-standard fields in the output of 'adjclust' (#13).
- Added tests for: equivalence with 'hclust', comparing sum of heights and pseudo inertia, plots, non-increasing heights, cutree (#14).
- Fixed #13 (man).
- Fixed #15 (Cutree with decreasing merges).
- Fixed #3 (Non-positive 'gains').
- Using BiocStyle::html_document2 as a temporary fix for vignette compilation errors.
Version 0.5.2 [2017-10-17]
- Added citation to Alia Dehman's PhD thesis to DESCRIPTION.
Version 0.5.1 [2017-10-16]
- More informative 'Description' of the method in DESCRIPTION
- Updates to test scripts to pass R CMD check on all windows platforms
- Moved README-*.png files to man/figures
Version 0.5.0 [2017-10-13]
- Bump version number for CRAN submission
Version 0.4.2 [2017-10-05]
- Added 'chac' S3 class and corresponding 'plot' and 'summary' methods
- Documentation cleanups
- Removed objects "R2.100" and "Dprime.100" (can be obtained from the imported 'snpStats' package)
- In 'snpClust': argument 'stat' is now passed to the 'snpStats::ld' function through '...'
- Some code cleanups
- Improved handling of default value for 'h' in 'adjclust' for 'dist' objects
- Renamed 'prevfit' into the more explicit 'res_adjclust_0.3.0'
- Dropped 'simmatrix' toy data set (now generated on the fly in tests)
Version 0.4.1 [2017-09-15]
- Cleanups in Hi-C and LD vignettes and corresponding tests
- Dropped outdated BALD test script
- Added test script for NA values in LD
- Renamed Hi-C data sets and updated corresponding documentation
- Added package website generated by pkgdown
Version 0.4.0 [2017-08-29]
- Implemented interface to handle standard and sparse matrices in adjClust
- Implemented interface to handle either kernel or dissimilarities
- Implemented wrapper for SNP and Hi-C data
- Documented the package and created vignettes for the different use cases
- Added scripts to increase package coverage and test the equivalence with rioja for the small dimensional case
- Cleaned up code to improve efficiency and removed unnecessary scripts and functions
Version 0.3.0 [2017-02-13]
- Removed 'adjClustBand': main entry points are now 'HeapHop' and 'adjClustBand_heap'.
- Updated test scripts and LD vignette accordingly.
- Added Travis CI and Appveyor support.
Version 0.2.*
Version 0.2.3 [2017-02-02]
- Updated LD vignette
- In adjClustBand, renamed flavor "Koskas" to "PseudoMatrix"
Version 0.2.2 [2016-12-01]
- Added dummy R/adjclust.R so that document() adds 'importFrom Rcpp evalCpp' to NAMESPACE
- "Fixed" warning at check due to .hpp file in src (this warning should not exist IMHO)
Version 0.2.1 [2016-11-09]
- Added minimal documentation
- Replaced "std::cout" by "Rcpp::Rcout", and so on for "exit()" and "cerr".
Version 0.2.0 [2016-06-24]
- Incorporated Michel's implementation (R function 'HeapHop')
- 'adjClustBand' is now a wrapper to call either Alia's or Michel's implementation
Version 0.1.0 [2016-06-24]
- Created from BALD
- Added a test to check that we are reproducing the results of BALD::cWard