repeatability: Estimate the repeatability of AFLP data
In AFLP: Analysis of AFLP data

Description Usage Arguments Details Value Author(s) See Also Examples

This function evaluates two indicators for the repeatability of the data: one based on the fluoresence and one on the classification. The indicators are based on all specimens with more than one replicate, outliers excluded. The indicators are given for both the specimens as the markers.

1
2
3

  repeatability(data, output = c("screen", "tex", "none"),
    bootstrap = FALSE, minMarker = NULL, path = NULL,
    device = "pdf")

`data`	An AFLP object with at least raw fluoresence data.
`output`	Which output is required. "screen" put graphics and possible outliers on the screen. "tex" givens the same information but saves the graphics to files and report LaTeX code to include the information in a LaTeX document. "none" suppresses the output.
`bootstrap`	Logical. Indicates whether a bootstrap procedure is run to detect possible outliers. These are specimens or markers which are probably unreliable because of a bad repeatability. Default to FALSE. Warning: setting this to TRUE can require a long computing time.
`minMarker`	All markers and specimens with a score less than `minMarker` are put in AFLP.outlier object. Defaults to NULL.
`path`	the path where the figures are saved. Only used if `output = "tex"`. Defaults to NULL, which is the working directory.
`device`	the device to which the figures are saved. See `ggsave` for the available devices. Only used if `output = "tex"`. Defaults to "pdf".

The indicator based on the fluorescence (raw or normalised) behaves like a variance. 0 equals a perfect match between all replicates from the same specimen. Higher values indicate less repeatable data. There is no upper limit for this value. The value is only useful to compare specimen or marker with the same project.

The indicator based on the score ranges from 0 (not reproducible at all, score is more or less random) to 1 (perfect repeatability).

The calculation of both indicator is described in the Details section.

First a selection is made from all possible combinations of marker and specimen were data of more than one replicate is available. This selection is used for both indicators.

The indicator on the fluorescence (raw and normalised) starts by calculating the variance of the fluorescence for each combination of specimen and marker. If the data is repeatable then the fluorescence will be very similar and hence the variance will be close to zero. The indicator per specimen is simply the mean of these variances over all markers. Likewise we the mean per marker of the variances over all specimens is the indicator per marker.

The indicator on the score is based on the number of possible mistakes. First the scores are converted into a binary score. The lowest class is considered 'absent', all other classed 'present'. Then we look at the number of 'absent' and 'present' replicates for each combination of marker and specimen. The class with the highest number is presumed to be the correct class. Hence the maximum number of mistakes for each combination of marker and specimen is the number of replicates divide by 2 and rounded downward. Now we have for all those combinations a number of mistakes and the maximum number of mistakes. We calculate for each specimen the sum of both numbers over all markers. Then we subtract the total number of mistakes from the total maximum number of mistake and divide that by the total maximum number of mistakes. If all replicates yield the same class, then no mistakes are made and the indicator equals 1. If the data has a very bad repeatability and half of the replicates are 'absent' and half 'present', then the total number of mistakes will equal the total maximum number of mistakes. This leads to an indicator equal to 0. The indicator per marker is calculated in a similar fashion (aggregation on marker instead of specimen).

data An ALFP object were the quality data is appended to.
Outliers An AFLP.outlier object with the possible outliers. This is based on the bootstrap procedure. Hence using boostrap = FALSE will result is an empty AFLP.outlier object.

Thierry Onkelinx Thierry.Onkelinx@inbo.be, Paul Quataert

normalise, classify, ggsave