RELAX-scan/RELAX.bf results

[francicco]

Hi @spond,

I'm running the Relax scan analysis. It seems like I always have different results using the same dataset and command line, also on the example data. I wonder if it is how it is suppose to be or there's something wrong with it. Cheers F

[spond]

Dear @francicco,

Some of these analyses (including RELAX-scan) are quite noisy (complex models). The original README was test-run on a HyPhy release that is quite old and the new version has more stable and robust optimization routines.

If you run hyphy RELAX.bf --alignment data/Fig4H.nex with the current version you get

1. A better log likelihood (-1614.99 vs -1616.21) for the 2-rate model
2. The same two branches flagged as possibly relaxed/intensified and two new branches, which were marginal in the previous version

Branch	Previous p	Current p	k (old/new)
Node 7	0.04	0.04	0 / 0.10
EU912347 ...	0.00	0.00	0.10 / 0.07
EU912347 ...	0.00	0.00	0.10 / 0.07
Node7	0.12	0.04	0.21 / 0.10
EU912349	0.09	0.01	0.22 / 0.11

Best, Sergei

[francicco]

Hi Sergei,

thanks a lot! What I'd to is run the regular RELAX on multiple subsequent branches and correct for multiple test afterward. It's probably slower but more reliable I guess.

Cheers F