Dear Yan,

1. In our experience, and depending on the amount of divergence, using REL on less than ten sequences can give some surprising results, to say the least, and having some closely related sequences with a few highly divergent sequences also appears to give misleading results. You may want to look into the estimates of the rate parameters that you get using REL; for small datasets, they will be subject to large estimation errors. In the 'Getting started' PDF that is included in the HyPhy distribution, there is a description of how you can get estimates of parameter error. If your dN/dS class is not significantly different from 1, then one cannot conclude with confidence that positive selection is present, even though the point estimate of dN/dS is greater than 1.

2. If you are looking for applications of REL in addition to those described in our paper 'Not so different...', we used it in our paper 'A Simple Hierarchical Approach to Modeling Distributions of Substitution Rates', linked to from Multimedia File Viewing and Clickable Links are available for Registered Members only!!  You need to Login . It was also used in a forthcoming paper in Virology by Pillai et al. comparing dN/dS and volatility as measures of selection in HIV.

Best wishes
Simon

Dear Yan,

All of the executables will give you exactly the same results; the difference between them is that they are compiled for different processors. For example, if you have a computer with a Pentium 4 processor, then using HYPHY-P4 will be faster, as it is compiled with optimizations specific for that processor, even though all of the programs run under Windows.

Best
Simon