Premium
Vergleich von unterschiedlichen Bootstrap‐Methoden zur Reduzierung der Verzerrung der QTL‐Effekt Schätzwerte
Author(s) -
Bennewitz J.,
Reinsch N.,
Kalm E.
Publication year - 2003
Publication title -
journal of animal breeding and genetics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.689
H-Index - 51
eISSN - 1439-0388
pISSN - 0931-2668
DOI - 10.1046/j.0931-2668.2003.00410.x
Subject(s) - statistics , mathematics , quantitative trait locus , parametric statistics , sampling bias , sample size determination , data set , calibration , biology , biochemistry , gene
Summary The factors leading to the bias of quantitative trait loci (QTL) effect estimates were investigated by simulating paternal half‐sib families. An upward bias was found in nearly all simulated configurations when only significant replicates were considered. This bias was a function of the experimental power of detecting a QTL (increased bias when power decreased) and was attributed to biased sampling. Three non‐parametric bootstrap schemes were tested to re‐estimate bias‐reduced QTL effects. The classical bootstrap bias corrected re‐estimate and the permutation bootstrap bias corrected re‐estimate failed to reduce the bias substantially. With the 0.368‐bootstrap the entire data set was repeatedly split into an estimation set formed by the bootstrap sample and into an independent test set formed by progeny not found in the estimation set. The size of the test set is asymptotically 0.368 times the number of progeny. The estimation set was used for calibration (i.e. estimating QTL position), and the test set was used for validation (i.e. estimating QTL effect at the particular position). The 0.368‐bootstrap re‐estimate was the mean effect estimate from validation from all bootstrap runs. The 0.368‐bootstrap produced on average significantly less biased QTL effects which showed reduced mean square errors compared with the original estimates. It was relatively robust against biased sampling.