If you are comparing two such frequencies, transform them to relative frequencies and a use t or z-test to compare them as proportions. If more than populations are to be compared you may transform them to relative frequencies and apply to them analysis of variance(ANOVA) and then if necessary multiple comparison test like Duncan's.

Dear Ette Etuk,

Thank you for your commentary.

I'm not entirely sure that you can do pairwise comparison between two values of mutation frequency. Each value represents a population that there is no associated mean or SEM. Likely I am wrong, but by now I wouldn't know how to perform that. 

Dear Jochen Wilhelm,

thank you for your answer.

Do you think is possible to do GzML comparing two ORFs of different lengths?

I would like to see if there are differences in the count of mutations between different genes.

Thanks

OK, I didn't know that it is possible with the loglinear function and assuming the Poisson distribution. I thought that using Poisson you have to use counts as dependent variable (as in "5 mutations and 995 non mutated nts in a ORF legnth of 1000 nts). But then it's also possible using mutations/nt or pb (mutation frequency).

Many thanks!

I'm thinking that GzLM using binary logistic should work as well, since it allows you to have non-mutated nuclotides data (from several clonal sequences for each ORF). I'm trying with counts of mutations as the dependent variable and total number of non-mutated nucleotides as the number of trials associated.

If using SPSS to do it, should I include intercepation in the model?

many times,  a warning appears at the beginning of the results window saying that hessian matrix is singular or that several values of the dependent variable are not integer or are less than zero. 

great, thanks!!

 Could anyone remind me why chi-square test is not proper for this analysis?

Apart from Chi-square tests,  you can use G2 tests.