create.diffsmatrix: Calculate Matrix of Distances between Peaks

Description

Generate an array of goodness-of-fit (or distance) between samples and knowns based on the sizes (in base pairs) of TRFLP peaks. For each sample/known combination, and for each enzyme/primer combination, this calculates the minimum distance between any peak in the sample and the single peak in the known.

Usage

create.diffsmatrix(samples, knowns)

Arguments

samples

A TRAMPsamples object, containing unidentified samples.

knowns

A TRAMPknowns object, containing identified TRFLP patterns.

Value

A three-dimensional matrix, with an attribute enzyme.primer, described above.

Details

This function will rarely need to be called directly, but does most of the calculations behind TRAMP, so it is useful to understand how this works. This function generates a three-dimensional $s \times k \times n$ matrix of the (smallest, see below) distance in base pairs between peaks in a collection of unknowns (run data) and a database of knowns for several enzyme/primer combinations. $s$ is the number of different samples in the samples data (length(labels(samples))), $k$ is the number of different types in the knowns database (length(labels(knowns))), and $n$ is the number of different enzyme/primer combinations. The enzyme/primer combinations used are all combinations present in the knowns database; combinations present only in the samples will be ignored. Not all samples need contain all enzyme/primer combinations present in the knowns. In the resulting array, m[i,j,k] is the difference (in base pairs) between the ith sample and the jth known for the kth enzyme/primer combination. The ordering of the $n$ enzyme/primer combinations is arbitrary, so a data.frame of combinations is included as the attribute enzyme.primer, where enzyme.primer$enzyme[k] and enzyme.primer$primer[k] correspond to enzyme and primer used for the distances in m[,,k]. Each case in the knowns database has a single (or no) peak for each enzyme/primer combination, but each sample may contain multiple peaks for an enzyme/primer combination; the difference is always the smallest distance from the sample to the known peak. Where a sample and/or a known lacks an enzyme/primer combination, the value of the difference is NA. The smallest absolute distance is taken between sample and known peaks, but the sign of the difference is preserved (negative where the closest sample peak was less than the known peak, positive where greater; see absolute.min).

Examples

Run this code

data(demo.samples)
data(demo.knowns)

s <- length(labels(demo.samples))
k <- length(labels(demo.knowns))
n <- nrow(unique(demo.knowns$data[c("enzyme", "primer")]))

m <- create.diffsmatrix(demo.samples, demo.knowns)

dim(m)
identical(dim(m), c(s, k, n))

## Maximum error for each sample/known (i.e. across all enzyme/primer
## combinations), similar to how calculated by \link{TRAMP}
error <- apply(abs(m), 1:2, max, na.rm=TRUE)
dim(error)

## Euclidian error (see ?\link{TRAMP})
error.euclid <- sqrt(rowSums(m^2, TRUE, 2))/rowSums(!is.na(m), dims=2)

## Euclidian and maximum error will require different values of
## accept.error in TRAMP:
plot(error, error.euclid, pch=".")