[Taxacom] Experimental TREEFINDER

[Gangolf Jobb]

A new experimental TREEFINDER version is online at:

www.treefinder.de

TREEFINDER is a software to compute phylogenetic trees from molecular
sequences.

New features are:

- general model selection and testing, using
- paired sites tests with information criterion
- parametric bootstrap test with information criterion
- output of sitewise information criteria
- two more information criteria: HQ and CAIC
- experimental simulated information criterion of my own: SimIC
- one more cost function for rate smoothing: NPRS-LOG

I would appreciate expert comments about the SimIC. This is an AIC-like
information criterion, but using an estimated "effective" number of
parameters as a penalty. The effective number of parameters is estimated
by parametric bootstrap as the gain in likelihood that is due to fitting
the noise of simulated data. It's described in the manual. For one thing
I am not sure whether the approach is new - sorry, I have here no access
to most of the literature. Second, the effective number of parameters often
appears to differ significantly from the true number of parameters. This
raises the question whether or not the traditional information criteria
such as AIC and BIC, when based on naive parameter counts, are applicable
in phylogenetics at all.

Please note that I am still not being paid for my work and that I had no
income for years.

Gangolf Jobb