Dear Sunil,
Your questions have just been pointed out to me - I'll try to answer the ones relating to PARRIS below.
Hope this helps,
Konrad Scheffler
---------------------
Quote:1. Would you recommend or discourage using parris as preliminary test for
selecting candidate genes for further site wise selection analysis?
Yes, that is the idea. PARRIS (the downloadable version for HyPhy) also does sitewise selection analysis, although the DataMonkey version does not.
Quote:2. Is there is way to know whether parris has detected and used any
recombination breakpoint in its analysis?
PARRIS does not scan for recombination - you need to feed it the recombination breakpoints as detected using a breakpoint detection program. Currently, the easiest way to do this is to use GARD and send its output to PARRIS. This is easy to do on DataMonkey, and the HyPhy version also accepts GARD output.
Quote: 3. I get different phylogenetic tree topology when same alligned CDS data is
analysed on datamonkey and on local hyphy(maximum likelihood
method>MG94>global>total branch length),both using NJ method? Though this is
said to have very little effect on results, i would like to know the reason
for this difference(does datamonkey use star phylogeny)?
I don't know - hopefully Sergei can answer this. I would be wary of incorrect topologies, after all recombination effectively causes false positives by messing up the topology.
Quote: 4. While running parris on datamonkey the selected nucleotide model changes in
the final step e.g. 012230 becomes 012240,010020 becomes 010040..why does this
happen?
Could be a bug - Sergei?
Quote:5. For parris result carried on datamonkey.. i consistently get p value of 1
and hence accepting the null hypothesis..LRT close to 0, both models having
almost same likelihood value..my 13 sequence alignments seem to be proper..
why does this happen?
This means that the alignment does not contain significant evidence for positive selection. I.e. after taking account of recombination breakpoints (if specified) and the fact that synonymous rates may vary (which can also be an artefact of recombination), there is no need to assume presence of positive selection in order to explain the data. This does not mean that positive selection is necessarily absent, but it does mean that you do not have enough evidence to infer its presence.
Quote:6. What is the setting for parris on hyphy to be same as one on datamonkey?ei
rate variation model = constant/proportional/dual/non-syn?
dual
Quote: it uses codon model or nucleotide model for branch length estimation?
I assume Sergei implemented it to use the nucleotide model.
Quote: whether the distribution is 'parris'?
I assume you are asking where in the distribution PARRIS is: it's in the folder Contributions/Possel_in_recomb/
Quote:7. Would you enligthen me a bit on analysing parris results in hyphy?
You need to be more specific with your question. PARRIS writes many results; hopefully some/most of them are self-explanatory.
Quote:now whether we are supposed to run LRT seperately on these models?(datamonkey does LRT by default..what about hyphy?) or just decide best fit based on AIC values??
The HyPhy script also does the LRT automatically - p-values are reported in the result files. If you prefer you can use the AIC instead - both methodologies should be valid in this case.