CHICKN_W1: Chromatogram Hierarchical Compressive K-means with Nystrom...
In chickn: 'Compressive' Hierarchical Kernel Clustering Toolbox

Description Usage Arguments Details Value References See Also Examples

An implementation of the complete pipeline of the CHICKN algorithm.

CHICKN_W1(
  Data,
  K = 2,
  k_total,
  K_W1 = NULL,
  kernel_type = "Gaussian",
  distance_type = "W1",
  Freq = NULL,
  ncores = 2,
  max_neighbors = 32,
  nblocks = 64,
  N0 = 10000,
  max_Nsize = 32,
  DoPreimage = FALSE,
  DIR_output = tempfile(),
  DIR_tmp = tempfile(),
  BIG = FALSE,
  verbose = FALSE,
  ...
)

`Data`	A Filebacked Big Matrix n x N.
`K`	Number of cluster at each call of clustering method. Default is 2.
`k_total`	An upper bound of the total number of clusters.
`K_W1`	A Filebacked Big Matrix. Nystrom kernel matrix s \times N, where N is the number of signals in the training collection and s is the Nystrom sample size. By default is NULL and it is generated using `Nystrom_kernel` function.
`kernel_type`	Kernel function type c('Gaussian', 'Laplacian').
`distance_type`	Distance function type. The available types are Wasserstein-1 ('W1') and Euclidean ('Euclide'). The default value is 'W1'.
`Freq`	A frequency matrix m x n with frequency vectors in rows. If NULL, the frequency vectors are generated by `GenerateFrequencies` function.
`ncores`	Number of cores. Default is 2.
`max_neighbors`	Number of neighbors used to estimate the kernel parameter `gamma`. Default is 32.
`nblocks`	Number of blocks, on which the regression is performed. Default is 32.
`N0`	Number of data vectors used for the variance estimation in `EstimSigma`.
`max_Nsize`	Number of neighbors used to compute consensus chromatograms.
`DoPreimage`	logical that controls whether to compute the consensus chromatograms. Default is TRUE.
`DIR_output`	A directory to save the results.
`DIR_tmp`	A directory for temporal files.
`BIG`	logical parameter that controls whether the resulting consensus chromatograms are stored as a Filebacked Big Matrix ('Centroid_preimage.bk'). Default is FALSE.
`verbose`	logical that indicates whether dysplay the processing steps.
`...`	Additional arguments passed on to `COMPR`.

CHICKN_W1 compresses the data by computing a Nystrom kernel approximation and applying the sketching operator from \insertCiteDBLP:journals/corr/KerivenBGP16chickn. See Nystrom_kernel and Sketch functions. Then clusters are recovered by operating on the compressed data version. It can use the kernel function based on the Wasserstein-1 or the Euclidean distances. It generates in DIR_output directory the following files:

'Cluster_assign_out.bk' is a Filebacked Big Matrix N x maxLevel+1, which stores the cluster assignment at each hierarchical level.
'Centroids_out.bk' is a Filebacked Big Matrix with the resulting cluster centroids in columns.

A list with the following attributes:

gamma is the estimated kernel parameter.
CompressedData is the Nystrom kernel matrix.
sigma is the estimated variance.
Frequency is the frequency matrix m x n.
Clusters is the cluster assignment.

Permiakova O, Guibert R, Kraut A, Fortin T, Hesse AM, Burger T (2020) "CHICKN: Extraction of peptide chromatographic elution profiles from large scale mass spectrometry data by means of Wasserstein compressive hierarchical cluster analysis." BMC Bioinformatics (under revision).

Nystrom_kernel, GenerateFrequencies, hcc_parallel, Preimage, bigstatsr

data("UPS2")
N = ncol(UPS2)
n= nrow(UPS2)
X_FBM = bigstatsr::FBM(init = UPS2, ncol=N, nrow = n)$save()
output  <- CHICKN_W1(Data = X_FBM, K = 2, k_total =8, max_neighbors = 10, ncores = 2, 
                     N0 = N, DoPreimage = FALSE)