BEAST 2 Help Me Choose

BEAST 2 Help Me Choose Frequencies prior

Frequencies prior

This is the prior on base frequencies for reversible nucleotide substitution models. Frequencies have to be in the range 0 to 1, and there is an extra constraint: the sum over all frequencies has to be 1. Usually, there is a sufficiently strong signal in alignment data for the prior on this parameter to be inconsequential.

A uniform [0,1] prior could be suitable, but in practice we know that it rarely happens that frequencies are very close to zero. A Dirichlet(4,4,4,4) prior still allows for a wide range of frequency values, but discourages values close to zero. This prior also performs well in well calibrated simulation studies (Bouckaert & Drummond, 2017).


Bouckaert RR, Drummond AJ. bModelTest: Bayesian phylogenetic site model averaging and model comparison. BMC evolutionary biology. 2017 Dec;17(1):1-1. doi:10.1186/s12862-017-0890-6.

Bayesian evolutionary analysis by sampling trees

Disclaimer: The above is the opinion of the author RB. If you do not agree, or spot a mistake, contact the author, or discuss this in the issues area or raise a new issue. A link will be added from this page to make sure others can find it.

Served through Jekyll, customised theme based on the twentyfourteen wordpress theme.