long branch and unrealible results in the runing

[Cloud LI]

   I used the PF2 results into iqtree, and it runs successfully, but I got an unreliable result compared to the results from CIPRES(ML-tree), 

a clade has a very long branch, and contains ten species that belong to some different clades/taxon in real. The parematers as follows:

 iqtree -s XXXX.phy -spp  XXXX.nex -bb 1000 

      And I also tried another method——run every gene tree in iqtree used the command line: iqtree -s XXXX.phy -m MFP, 

after that, I got the best fit model for every gene used BIC, and I write a NEXUS format file manually like this:

#nexus

begin sets;

charset part1 = 1 -1530;

charset part2 = 1531 -2112;

charset part3 = 2113 -3114;

.......

charpartition mine = TVMe+R3:part1 , K3Pu+F+R2:part2 .......

At last, I used :

 iqtree -s XXXX.phy -spp  XXXX.nex -m MFP -bb 1000

There are warnning messages: 

 WARNING: Too high (saturated) partition rates for proportional partition model!

Current log-likelihood at step 1: -192589.529

But actually, I used plant mitochondrial genes, so generally, all of the genes have a relative lower evolutioan rate.

    I was very confused if I used a set of wrong parameters or if I had misunderstood these procedures.

I  look forward to your reply. 

Thanks.

Yunxia