Skip to main content
ARS Home » Southeast Area » Raleigh, North Carolina » Plant Science Research » Research » Publications at this Location » Publication #184591
Title: MAPPING WHEN PHENOTYPE MEASUREMENTS ARE NOT WELL BEHAVED: COMPARISON OF RECUSIVE PARITIONING WITH COMPOSITE INTERVAL QTL MAPPING

Author
item STAPLETON, ANN - UNC-WILMINGTON
item SIMMONS, SUSAN - UNC-WILMINGTON
item ROBERTSON, LEILANI - NORTH CAROLINA STATE UNIV
item Holland, Jim - Jim

Submitted to: Maize Genetics Conference Abstracts
Publication Type: Abstract Only
Publication Acceptance Date: 1/3/2005
Publication Date: 3/10/2005
Citation: Stapleton, A., Simmons, S., Robertson, L., Holland, J.B. 2005. Mapping when phenotype measurements are not well behaved: comparison of recusive paritioning with composite interval QTL mapping [abstract]. Maize Genetics Conference.

Interpretive Summary: We developed and tested a method for mapping the locations of genes affecting complex traits that are expressed atypical ways. Some traits, like ear number, do not follow a bell-curve (normal) distribution in maize populations, but are expressed as either 0, 1, or 2 ears. Typical gene mapping methods do not account for such unusual expression patterns. Recursive partitioning, a new method, was applied to unusually distributed data and found to identify the underlying genes with good accuracy.

Technical Abstract: Standard methods for QTL analysis use a summary measure such as the mean of the trait measurement from each line. The phentoype measurements are assumed to be normally distributed. In practice, a good fit to a normal distribution is rare, and transformations are often attempted to correct the distribution. This complicates interpretation of the QTLs. Recursive partitioning is a powerful algorithm that investigates the association among various predictor variables and a response (or multiple responses). Some advantages recursive partitioning offers over other conventional algorithms are 1) it does not assume the relationship between the response variable and the predictor (marker) variables is linear; 2) missing values are easily implemented into the analysis without the problem of imputing values; 3) it can handle data sets where there are many more predictors, p, than sample size, n; 4) associations among predictor variables are not an issue, in fact, recursive partitioning will uncover the associations and interactions between the various predictor variables; 5) in the case of QTL analysis, the number of QTLs need not be known apriori nor any assumptions made regarding interactions between QTLs. We compare recursive partitioning and regression-based methods using simulations. We then compare analyses of data on Fusarium ear rot susceptibility in an BC1F2 and an RIL population, and ear fill in one RIL data set, using recursive partitioning, single-marker ANOVA, and composite interval mapping. We develop guidelines for use of recursive partitioning versus the current standard, composite interval mapping. Recursive partitioning is generally applicable for genome scanning; the method is most useful for situations in which 1) there is no map available (as in polyploid species or early in genotyping efforts), 2) the trait data have non-normal distributions and/or the trait data are poorly described by summary measures such as means, and 3) in mapping populations that have high marker density or have small numbers of lines such that the number of lines is less than the number of markers.