Discrepancy between Branch model M2 and Clade model C

[x...@ualberta.ca]

Hi everyone,

I identified my gene of interest from 24 different plant species using BLAST. In the end ,I have 70 cDNA sequences in total. These sequences can be divided into three distinct clades with high bootstrap support based on both MrBayes and RAxML analysis. Some species contain genes in all three clades. Some only contain genes in one or two clades. The objective of my work is to detect the functional divergence among three clades (named subgroup I, II and III)

First I ran the M2 branch model (M2 NS0) using each subgroup linage as the foreground branch (labeled as #1 in tree file). The LRT showed that setting the subgroup I lineage (but not Subgoup II and III) as the foreground branch fits the data significantly better than the one-ratio M0 model. In addition, Subgroup I has a lower rate of dN/dS (w) value as compared with the subgroup II and III.

Then I fit the my data to the MC model (M3 NS2). Setting subgroup I clade as the foreground partition ( labeled as $1 in tree file) provided a significant better fit than null model M2a_rel. Setting subgroup II (but not Subgroup III) also resulted better fit. However, in contrast to branch model result, subgroup I has a higher w value than subgroup II and III.
My question is what might be the reason for this discrepancy? If Subgroup I is under relaxed constraints, should it has higher values in both branch and clade model analysis?

Here are some information related to my control file:
1. Tree file: The paralogous gene from yeast was used as the ourgroup for phylogenetic tree analysis. The unrooted tree without branch length was used for analysis.
2. Sequence file: The sequences were aligned based on amino acid sequence and cleaned by Gblock before back-translated to DNA. There are some gaps in the sequences and I chose to keep sites with ambiguity data.
3. Codon Freq: F3x4
4. I tried different initial omega value (0.1, 0.4 and 1) and the results are consistent.
5. Clock was set as 0
6. ncatG was set as 8


Your help is highly appreciated!!!

Thanks in advance,

Xue

[x...@ualberta.ca]

Hi everyone,

So I tried using different codon frequencies for the branch model M2 analysis(M2 NS0). When I used F1/16 mode, the Subgroup I has a higher rate of dN/dS (w) value as compared with the subgroup II and III, which is consistent with Clade model analysis. However, the LRT tests indicated that M2 model does not fit the data significantly better than the one-ratio M0 model. So my question is which codon frequency is better for branch model analysis and which result is the correct one?

Thanks,
Xue

[x...@ualberta.ca]

Hi again,

I continued to try different settings for the branch model analysis. One thing I found is that instead of just labeling the single leading branch of the clade, labeling the branches in the whole clade (basically I put the labeling sign at the same place in the tree file, but change "#" to "$"）give me consistent result ( Sub I has higher w ratio than SubII and SubIII and M2 model provides better fit than M0 model) in both codon frequency model (F1/16 and F3X4). This result is also consistent with clade model analysis.

So is this the right way to label the branches in M2 model (the whole clade instead of just the single branch)?

Any suggestions would be appreciated,

Thanks in advance,

Xue

[Ziheng]

it seems that you have resolved the issue yourself.
in the branch model, you labeled one single branch, while in the clade model you labeled all branches within the whole clade. Why do you expect to see similar results between such two drastically different models?
Anyway the difference between the two symbols for labelling branches is explained in the doc: pamlDOC.pdf.
ziheng