I have a set of HIV pol sequence data, and after alignment it is clear that some sequences are much shorter than others – primarily due to sequencing issues. Because the gaps/missing data are not necessarily meaningful, I consider these a "complete deletion" when doing a maximum likelihood phylogenetic analysis. Doing this, however, reduces the number of positions that could be included in the analysis since only positions where there is data for all sequences can be used. I was wondering what the best approach to handle this would be. Is there a general lower limit for the number of bases that should be included for an informative viral phylogenetic analysis, and is it better to remove samples with shorter sequence lengths? Would a Bayesian approach using programs like BEAST overcome this? Thank you!
There is no set answer to this type of question. It depends on what type of organism you are studying, and even which gene within a given type of organism. It also depends on what information you want from the phylogeny. For HIV-1 M group viruses, for example, the pol gene has less phylogenetic signal per 100 bases than the env gene does because pol is under much more purifying selection pressure to retain functions (protease, RT, integrase, RNAse etc) and less selection pressure from the host immune system.
If you only need to know what subtype the virus is, a few hundred bases of the pol gene can be enough, but if you want to infer transmission networks you will need more data. Recombination and other issues also create problems. Many HIV researchers now create database entries of two regions of pol (protease and the core of RT) spliced together without noting that there is missing data between the two regions.
There is no set answer to this type of question. It depends on what type of organism you are studying, and even which gene within a given type of organism. It also depends on what information you want from the phylogeny. For HIV-1 M group viruses, for example, the pol gene has less phylogenetic signal per 100 bases than the env gene does because pol is under much more purifying selection pressure to retain functions (protease, RT, integrase, RNAse etc) and less selection pressure from the host immune system.
If you only need to know what subtype the virus is, a few hundred bases of the pol gene can be enough, but if you want to infer transmission networks you will need more data. Recombination and other issues also create problems. Many HIV researchers now create database entries of two regions of pol (protease and the core of RT) spliced together without noting that there is missing data between the two regions.
Thank you Brian Thomas Foley this is very helpful, and I appreciate the response!
Overall, I am not a huge fan of bootstrap values, such as implying that bootstrap support of 70% or 90% or 99% for some subclade is significant. However I do think that observing whether or not you can get high bootstrap values from your data set is one good indication of whether or not you have enough data to make the points you want to make. Low bootstrap support may not always be overcome with adding length to the alignment, for example when recombination has created more of a "web" than a "tree". But anyway, bootstrapping is built into most phylogenetic packages (PHYLIP, PAUP, MEGA, DAMBE, etc) so it is easy to do. Even if you want your final tree for publication to be made with the best maximum likelihood or Bayesian method using all of your data, it can be informative to check some subsets and see for example if using half and 3/4 of the data gives close to the same result as using all the data. In my experience, there is no substitute for empirically testing your data set with different methods and/or different subsamples of the data.
- Badges
- Science method
- Similar topics
- Biological Science
- Evolutionary Biology
- Phylogenetics
- Phylogenetic Analysis