Dear All,

A simple trick to speedup the pruning algorithm to compute tree likelihood exploiting the time-reversible property of Markov models is described in the supplementary material of this paper in MBE: https://doi.org/10.1093/molbev/msx281. This trick was implemented in RAxML and IQ-TREE. The gist is to speed up branch length estimation, which dominates the runtime. The runtime reduction is 4, 20 and 61 times for DNA, protein and codon models, respectively.

Regards, Minh