Validation and power of the Mixture Model

Dear all, I am a fresh research student. Previously, I had the experience in developing a pharmacogenetics-based population PK model using NONMEM, but this is the first time I work with the Mixture Model in NONMEM and I have some questions about it. I have two datasets on hand, where patients' clearance are believed to be affected by their genotypes. While I have their genotypic information, I also wish to know if the Mixture Model in NONMEM can help me accurately categorize the population even if the genotypic information are "hidden" from NONMEM. The results turned out to be unsatisfactory somehow. For the subpopulations with a distinctively different typical values of clearance, the sensitivity and specificity can approach 100%, but for those with less differences, the average accuracy drops to 60-70%. Although it is not difficult to understand that the computer will not be able to categorize these subjects when they have similar parameters (either mean values too close or variances too large...), I am wondering if there is any general approach to utilize the best out of the Mixture Model function. Regarding the power of the Mixture Model, I wonder if there has been any validation done before for datasets with different characteristics. For examples, is there any previous study that looked into the accuracy of the Mixture Model function and can somehow express the typical accuracy in terms of the difference in, say, the mean plasma levels, between 2 subpopulations. Last but not least, it would be great if anyone can kindly advise me any good teaching materials about the Mixture Model in NONMEM. Sincerely, Matthew Hui