Plant Genome IV Conference Plant Genome IV Conference
Maximum likelihood approach has been said the method of choice for gene ordering. However, questions for the likelihood approach using in gene ordering remain. When a perfect model is used for a mapping data set, the log likelihoods should be same for all possible locus orders for the same data set. The general argument to support this point will be that parameters for a fully parameterized model are sufficient statistics to explain all the variation in the data. For gene ordering problem, if the parameters for the data are sufficient and the maximum likelihood estimates of the parameters are unbiased for different orders, the log likelihoods evaluated using the maximum likelihood estimates should be same for all possible gene orders. Another argument against the likelihood approach is that the likelihood comparison cannot be implemented as the formerly log likelihood ratio test because there is no degree of freedom difference for the parameters among the possible locus orders. The parameter spaces are same for all possible locus orders. Restrained likelihoods (assuming absence of or complete interference) have shown usefulness for determining the most possible locus order for three-locus situations. Simulation results on ML gene ordering will be reported in this poster.