Skip to contents

Targeted analysis

Source: R/targeted_analysis.R
targeted_analysis.Rd

Targeted analysis measures a set of known compounds. Two routes lead from a table of target compounds to a peak table with a named column for each target: selecting peaks by retention time in get_peaks, or matching peaks to a reference sample with match_peaks.

Choosing a route

Select peaks by retention time using get_peaks when the samples are aligned, or can be aligned with correct_rt, and no other peak falls within tol of a target. Only the peaks near the targets are fitted, which is quicker, and each target can be measured at its own wavelength.

Match peaks to a reference using match_peaks when retention times alone cannot identify the targets (e.g., in crowded windows). Peaks are matched by spectrum as well as retention time, so a target can be picked out from its neighbors, and the retention-time window can be wider to allow for drift. Every peak is fitted, and every target is measured at the same wavelength.

Constructing the targets table

Build a data.frame with an rt column giving the retention time of each target and, optionally:

  • name: Unique names, used as peak table column names.

  • lambda: The wavelength at which to measure each target. When matching to a reference, it must equal the lambda given to match_peaks.

  • tol: How far each target's peak may lie from its rt, in minutes, regardless of the time unit of the chromatograms. Defaults to 0.1.

  • spectral_cutoff: Used only by match_peaks. The minimum spectral correlation for a match to each target's reference peak. Defaults to the spectral_cutoff argument of match_peaks.

  • rt_window: Used only by match_peaks. The maximum difference in retention time, in minutes, between each target's reference peak and its match in another sample. Defaults to the rt_window argument of match_peaks.

Selecting peaks by retention time

  1. Call get_peaks(chrom_list, targets = targets). In each sample, get_peaks fits only the peaks whose apex lies within tol of a target.

  2. Call get_peaktable on the result. The peak table has one column per target found in at least one sample; a target found in no sample gets no column.

Peaks are assigned by retention time only, so each target must stay within tol of its listed time in every sample, and closer to it than to any neighboring target. If the samples drift more than that, fit warping models with correct_rt and pass them as a mod_list; the targets are then found on the aligned time axis.

When more than one peak falls within tol of a target, the select_by argument of get_peaks decides which is retained. The default, select_by = "height", keeps the tallest, which may not be the target; narrow tol, or set select_by = "proximity" to keep the peak closest to the target. Peaks are selected before they are fitted, so "height" compares the signal at each apex, including any baseline beneath it.

To keep every candidate in the peak list, set select_by = "none". Where a sample has several peaks for one target, get_peaktable enters the one with the largest response in the table, but summarizes the target's pk_meta over all of them, so its rt can fall between the candidates. To take the metadata from the most intense peak instead, set summarize_by = "max" in get_peaktable.

Matching peaks to a reference sample

  1. Call get_peaks without providing targets, so that every peak is fitted.

  2. Call match_peaks(peak_list, targets = targets, lambda = ...). Unless reference is supplied, match_peaks picks as reference a sample that has a peak for every target, then matches each sample's peaks to the reference peaks by spectral correlation and retention time.

  3. Call get_peaktable(matched, pks = peak_list). Unlike the other pathway, the peak table will still include untargeted peaks. Use [, targets$name] to keep only the targets.

In the reference sample, each target is assigned to the tallest well-fitted peak within tol minutes of the specified rt. A peak in another sample can match a reference peak only if it lies within rt_window minutes of it and their spectral correlation is above the spectral_cutoff.

See also

get_peaks, match_peaks, get_peaktable

Author

Ethan Bass

Examples

data(Sa_pr)
targets <- data.frame(name = c("A", "B"), rt = c(12.07, 14.31))

# selecting by retention time, on the aligned time axis
mods <- correct_rt(Sa_pr, lambdas = 220, what = "models", reference = 1)
pks_t <- get_peaks(Sa_pr, lambdas = 220, targets = targets, mod_list = mods)
get_peaktable(pks_t)
#>             A        B
#> 119  97.99150 40.10133
#> 121  69.62984 26.52917
#> 122  72.12245 34.11141
#> 458 134.66248 29.94042

# matching to a reference sample
pks <- get_peaks(Sa_pr, lambdas = 220)
matched <- match_peaks(pks, targets = targets, lambda = 220)
get_peaktable(matched, pks = pks)[, targets$name]
#>             A        B
#> 119  97.99150 40.10133
#> 121  69.62984 26.52917
#> 122  72.12245 34.11141
#> 458 134.66248 29.94042