| In the first chapter, some issues about detecting positive selection within protein-coding genes are addressed. Codon models are introduced, including the random effect models and the fixed effect models. The LRT is illustrated on the nested random effect models and fixed effect models for testing hypotheses of interest.;In the second chapter, the generalized fixed-effect model, which includes a "random effect step" and a "fixed effect step", is developed to enhance the applicability of the fixed effect models. We also extend the 5 existing fixed effect models to a full range of 16 models which would provide a greater flexibility to investigate the heterogeneous evolution among sites in a protein. LRT and AIC are used to select the best fixed effect model for real genes.;In the third chapter, a new method called the Likelihood Based Clustering method (LiBaC) is implemented to modify the a priori partition of the fixed effect models. Simulated data are used to check the performance of LiBaC based on: (i) error rates of site partition; (ii) likelihood values; (iii) parameter estimates. An example of its application in real gene analysis is shown in the end.;The conclusions and some directions for future work are addressed in the fourth chapter. |
