ChemoSpec (version 4.0.1)

corSpectra: Tools for STOCSY Analysis of a Spectra Object

Description

These functions provide tools for STOSCY analysis of a Spectra object, using the method developed by Nicholson. STOCSY is Statistical Total Correlation Spectroscopy. Briefly, the correlation matrix of an NMR data set of n samples and p frequencies is computed (the matrix dimensions are p x p). Peaks arising from the same compound are intrinsically positively correlated. This is much like a 1D or 2D TOCSY NMR spectrum. However, peaks that are correlated, positively or negatively, due to metabolic processes, will also appear in the STOCSY plot. corSpectra computes the correlation and covariance matrices, and can display the correlation matrix in several formats. Detailed inspection/interpretation of this plot is tedious, and producing it can be slow for large data sets, so it's most useful as an overview. covSpectra will display a single frequency (i.e. chemical shift) from the covariance matrix, but color it according to the correlation matrix values. This is point where detailed interpretation is done. See the example.

Usage

corSpectra(spectra, plot = TRUE, limX = NULL, limY = NULL, nticks = 10,
levels = NULL, pmode = "contour", C = NULL, V = NULL, ...)
covSpectra(spectra, freq = spectra$freq[1], C = NULL, V = NULL, ...)

Arguments

spectra

An object of S3 class Spectra.

plot

Logical. Should a plot be made? Applies to corSpectra only.

limX

Numeric vector of length 2. The x limits. Applies to corSpectra only.

limY

Numeric vector of length 2. The y limits. Applies to corSpectra only.

nticks

Integer. The number of ticks to be drawn. Applies to corSpectra only.

levels

Numeric. A vector of values at which to draw the contours or levels. Applies to corSpectra only. The default is to use chooseLvls to compute 5 evenly spaced levels. For most data sets th

pmode

Character. The plot mode. Applies to corSpectra only. One of c("contour", "image", "contourplot", "levelplot", "rgl", "exCon"). The last two are interactive. See Details.

C

Matrix. Optional. A precomputed correlation matrix.

V

Matrix. Optional. A precomputed covariance matrix.

freq

Numeric. The desired frequency to be plotted. Applies to covSpectra only.

...

Other parameters to be passed to the plotting functions.

Value

• A list giving the covariance and correlation matrices. A plot will also be made if plot = TRUE for corSpectra.

Warning

For corSpectra, the labeling of the axis will be wrong if there is a gap in the data.

Details

If spectra$freq is in decreasing order, it and spectra$data are silently re-ordered to be increasing before plotting.

The calculation of the correlation and covariance matrices may take quite some time for large data sets. It is possible to pre-compute these and pass them into the functions to save time and avoid repetition.

Plotting in corSpectra can be extremely slow for large data sets. The base graphics options (pmode = "contour" or "image") are much faster than the lattice options. These plots are probably best for an overall sense of the data and for publication rather than detailed interpretation. If using pmode = "contour" drawing fewer contours is of course faster for both drawing and computation of the contours. Note too that contour style plots have n colors for n contour levels but image style plots have n-1 colors for n levels.

The color scale for the plots is blue/low correlation to red/high correlation, anchored at a shade of green for zero correlation. The example shows how to see the color scale.

For covSpectra, the x and y limits can be set simply by passing xlim and ylim via the ....

References

O. Cloarec et. al. Analytical Chemistry vol. 77 pgs. 1282-1289 (2005).

https://github.com/bryanhanson/ChemoSpec

See Also

chooseLvls to select levels automatically.

getMaxCovByFreq is a utility to sort the covariance matrix by the maximum covariance at a given frequency. This can be used to identify the strongest interactions.

Examples

# The color scale used:
# cscale <- c(rev(rainbow(4, start = 0.45, end = 0.66)), rev(rainbow(5, start = 0.0, end = 0.25)))
# pie(rep(1, 9), col = cscale)
#
# This data set is a mixture of compounds, some of
# which have correlated concentrations, both
# positively and negatively.  The contour plot shows
# correlations w/i a spin system and among compounds.
# The 2nd plot shows one particular frequency.
# For more info about the data set see ?MUD
#
data(metMUD2)
lvs <- c(-0.99, -0.95, -0.9, 0.9, 0.95, 0.99)
lim <- c(0.6, 4.3)
res <- corSpectra(metMUD2, levels = lvs, limX = lim, limY = lim, main = "metMUD2 NMR Data")
jnk <- covSpectra(metMUD2, freq = 1.0, C = res[[2]], V = res[[1]])