cluster_spectra: Cluster peaks by spectral similarity.

In chromatographR: Chromatographic Data Analysis Toolset

Cluster peaks by spectral similarity.

Description

Function to cluster peaks by spectral similarity. A representative spectrum is selected for each peak in the provided peak table and used to construct a distance matrix based on spectral similarity (pearson correlation) between peaks. Hierarchical clustering with bootstrap resampling is performed on the resulting correlation matrix to classify peaks into by their spectral similarity.

Usage

cluster_spectra(
  peak_table,
  chrom_list,
  peak_no = c(5, 100),
  alpha = 0.95,
  nboot = 1000,
  plot_dend = TRUE,
  plot_spectra = TRUE,
  verbose = TRUE,
  save = TRUE,
  parallel = TRUE,
  max.only = FALSE,
  output = c("clusters", "pvclust", "both"),
  ...
)

Arguments

peak_table	Peak table from get_peaktable.
chrom_list	A list of chromatograms in matrix form (timepoints x wavelengths).
peak_no	Minimum and maximum thresholds for the number of peaks a cluster may have.
alpha	Confidence threshold for inclusion of cluster.
nboot	Number of bootstrap replicates for pvclust.
plot_dend	Logical. If TRUE, plots dendrogram with bootstrap values.
plot_spectra	Logical. If TRUE, plots overlapping spectra for each cluster.
verbose	Logical. If TRUE, prints progress report to console.
save	Logical. If TRUE, saves pvclust object to current directory.
parallel	Logical. If TRUE, use parallel processing for pvclust.
max.only	Logical. If TRUE, returns only highest level for nested dendrograms.
output	What to return. Either clusters to return list of clusters, pvclust to return pvclust object, or both to return both items.
...	Additional arguments to pvclust.

Details

A representative spectrum is selected for each peak in the provided peak table and used to construct a distance matrix based on spectral similarity (pearson correlation) between peaks. It is suggested to attach representative spectra to the peak_table using attach_ref_spectra. Otherwise, representative spectra are obtained from the chromatogram with the highest absorbance at lambda max.

Hierarchical clustering with bootstrap resampling is performed on the resulting correlation matrix, as implemented in pvclust. Finally, bootstrap values can be used to select clusters that exceed a certain confidence threshold as defined by alpha. Clusters can also be filtered by the minimum and maximum size of the cluster using the argument peak_no. If max_only is TRUE, only the largest cluster in a nested dendrogram of clusters meeting the confidence threshold will be returned.

Value

Returns clusters and/or pvclust object according to the value of the output argument.

• If output = clusters, returns a list of S4 cluster objects.

• If output = pvclust, returns a pvclust object.

• If output = both, returns a nested list containing [[1]] the pvclust object, and [[2]] the list of S4 cluster objects.

The cluster objects consist of the following components:

• peaks: a character vector containing the names of all peaks contained in the given cluster.

• pval: a numeric vector of length 1 containing the bootstrap p-value (au) for the given cluster.

Note

• Users should be aware that the clustering algorithm will often return nested clusters. Thus, an individual peak could appear in more than one cluster.

• It is highly suggested to use more than 100 bootstraps if you run the clustering algorithm on real data even though we use nboot = 100 in the example to reduce runtime. The authors of pvclust suggest nboot = 10000.

Author(s)

Ethan Bass

References

R. Suzuki & H. Shimodaira. 2006. Pvclust: an R package for assessing the uncertainty in hierarchical clustering. Bioinformatics, 22(12):1540-1542. doi: 10.1093/bioinformatics/btl117.

Examples

data(pk_tab)
data(Sa_warp)
cl <- cluster_spectra(pk_tab, nboot=100, max.only = FALSE, save = FALSE, alpha = .97)

chromatographR documentation built on Aug. 24, 2022, 9:06 a.m.