RAxML BlackBox
Model specification
This allows you to specify the regions of your alignment for which an individual 
model of nucleotide substitution should be estimated. This will typically 
be useful to infer trees for long (in terms of base-pairs) multi-gene alignments. 
Individual α-shape parameters, GTR-rates, and 
base frequencies will be estimated and optimized for each partition. 

IMPORTANT CHANGE w.r.t. previous versions:
Since RAxML can now handle mixed DNA and AA alignments you MUST specify the type of data in the partition file, before the partition name. For DNA data this just means that you have to add DNA to each line in the partition file, for AA data this is done by specifying the respective AA substitution matrix you want to use for a partition. If you want to do a mixed/partitioned analysis of a concatenated AA and DNA alignment you can either specify -m GTRCAT or, e.g., -m PROTCATWAG, the only thing that will be -extracted from the string passed via m is the model of rate heterogeneity you want to use.

If you 
have an alignment with 1,000bp from two genes gene1 (positions 1-500) and 
gene2 (positions 501-1,000) the information in the multiple model file should 
look as follows: 

gene1 = 1-500
gene2 = 501-1000

If gene1 is scattered through the alignment, e.g. positions 1-200, and 
800-1,000 you specify this with: 

gene1 = 1-200, 800-1,000
gene2 = 201-799

You can also assign distinct models to the codon positions, i.e. if you want 
a distinct model to be estimated for each codon position in gene1 you can 
specify: 

gene1codon1 = 1-500\3
gene1codon2 = 2-500\3
gene1codon3 = 3-500\3
gene2 = 501-1000

If you only need a distinct model for the 3rd codon position you can write: 

gene1codon1andcodon2 = 1-500\3, 2-500\3 

gene1codon3 = 3-500\3 

gene2 = 501-1000 

Finally, only for AA data its is also possible to specified transition 
matrices for each partition: 

JTT, gene1 = 1-500 

WAGF, gene2 = 501-800 

gene3 = 801-1000 

The AA substitution model must be the first entry in each line and must 
be separated by a comma from the gene name. If like in the example you do 
not specify a matrix for gene3 the model for gene3 will default to the model 
specified in the substitution matrix section. 
When publishing results obtained via the web-server please cite:
A. Stamatakis, P. Hoover, J. Rougemont
A Rapid Bootstrap Algorithm for the RAxML Web-Servers,
Systematic Biology, 75(5): 758-771, 2008