Post by Sergei on Feb 1st, 2012 at 12:19pm
Hi Lun,
For 9 sequences regardless of the method there is generally very little power to say anything about any particular sites. You can use BranchSiteREL to test whether or not a proportion of sites at a given lineage have been subjected to selection, but you won't have any confidence that any particular site is selected. You can run MEME on datamonkey.org if you are determined to try to find specific sites under selection: this method extends FEL to deal with heterotachy (as suggested by Konrad). Generally, I would advise you AGAINST reporting any specific sites under selection when you have 9 closely related sequences -- you should try to make statements about proportions of sites (BranchSiteREL or REL), e.g. there is evidence that X% of sites are evolving with dN/dS > 1. You can also supply a plot of site-wise dN/dS estimates (which datamonkey outputs), e.g. by REL, to indicate trends of spatial localization.
The quote from the SARS paper can be a bit misleading -- REL can pool signal from multiple sites to boost power to detect selection somewhere in the gene, but you still won't get great precision/power at any given site (see Multimedia File Viewing and Clickable Links are available for Registered Members only!! You need to
, page 26):
The REL approach can be used to test whether positive selection operated on a proportion of sites in an alignment
As far as GARD is concerned, please perform the analysis with all 4 species, but don't include too many sequences per species. If you think there is no INTRA-sepcies recombination, I would include one sequence from A,B,C and D. See Multimedia File Viewing and Clickable Links are available for Registered Members only!! You need to page 8.
Sergei
For 9 sequences regardless of the method there is generally very little power to say anything about any particular sites. You can use BranchSiteREL to test whether or not a proportion of sites at a given lineage have been subjected to selection, but you won't have any confidence that any particular site is selected. You can run MEME on datamonkey.org if you are determined to try to find specific sites under selection: this method extends FEL to deal with heterotachy (as suggested by Konrad). Generally, I would advise you AGAINST reporting any specific sites under selection when you have 9 closely related sequences -- you should try to make statements about proportions of sites (BranchSiteREL or REL), e.g. there is evidence that X% of sites are evolving with dN/dS > 1. You can also supply a plot of site-wise dN/dS estimates (which datamonkey outputs), e.g. by REL, to indicate trends of spatial localization.
The quote from the SARS paper can be a bit misleading -- REL can pool signal from multiple sites to boost power to detect selection somewhere in the gene, but you still won't get great precision/power at any given site (see Multimedia File Viewing and Clickable Links are available for Registered Members only!! You need to
, page 26):The REL approach can be used to test whether positive selection operated on a proportion of sites in an alignment
As far as GARD is concerned, please perform the analysis with all 4 species, but don't include too many sequences per species. If you think there is no INTRA-sepcies recombination, I would include one sequence from A,B,C and D. See Multimedia File Viewing and Clickable Links are available for Registered Members only!! You need to
page 8.Sergei
http://www.hyphy.org/cgi-bin/hyphy_forums/YaBB.pl?action=downloadfile;file=BSR.pdf (141 KB | )