Summary: | 碩士 === 國立臺灣大學 === 農藝學系研究所 === 86 === Most quantitative traits are assumed to be continuously and
normally distributed.
Due to the nature of response, limitation of measurement or
some other theoretical or
practical considerations, only the discrete binary data are
available. Published statistical
methods for mapping and analyzing quantitative trait loci(QTL)
all center on the normal
assumption and thus can not be directly applied for binary
distributed data. This study
proposes a logistic mixture regression model for binary
distributed data based on simple
interval mapping(SIM) and composite interval mapping(CIM)
respectively. Iteratively
reweighted least squares(IRLS) derived by expectation-
maximization(EM) algorithm is
used to obtain the maximum likelihood solutions of the effects
and positions of QTLs. The
methods presented in this paper are conceptually simple, easy
to implement and fast in
convergency. Results from simulated F2 intercross data indicate
the proposed methods
can effectively detect the putative QTLs. Method based on SIM
has higher power than
method based on CIM for the case of single QTL on each
chromosome. Method based on
SIM becomes less effective than method based on CIM when there
are two or more QTLs
on each chromosome, especially when QTLs are close together. In
other words, method
based on CIM has higher resolution than method based on SIM
when there are two or
more QTLs on each chromosome. However, interpretation of
genetic parameters is more
difficult for CIM model than for SIM model. Since number of
QTLs on each chromosome
is not known in practice, both CIM and SIM models should be
tested and compared for
any particular set of data. Two linked QTLs are not likely to
be distinguished by both the
methods studied if the distance between QTLs is less than about 20cM.
|