Hypermutation is a phenomena that affects HIV-1 introducing large numbers of mutations into some sequences. It manifests in the datasets as sequences in which large numbers of Guanine was mutated to Adenine, specifically when that Guanine was surrounded by a particular pattern. The hypermut 2.0 tool available from https://www.hiv.lanl.gov/content/sequence/HYPERMUT/hypermut.html is a frequently used tool to detect and remove hypermutated sequences. We wrote a new implementation of the hypermut 2.0 algorithm in R.
The hypermut algorithm compares each sequence in an alignment to some ancestral sequence (usually approximated by the consensus sequence of the alignment), tallying the frequency of specific mutations. Hypermutation occurs when a G which is followed by an A or G (denoted by R in the IUPAC convention) and then by an A, G or T (denoted by a D in the IUPAC convention) mutates to an A. More compactly, when GRD become ARD, the mutation is flagged as possibly due to hypermutation. In order to distinguish between true hypermutation and the generally expected level of mutation, a baseline must be established. The baseline is established by tallying G to A mutations when the G is followed immediately by either a C or T (denoted by a Y in the IUPAC convention) or when the G is followed by an A or a G (denoted by R in the IUPAC convention) and then a C. More compactly, when GY becomes AY or GRC becomes ARC, the mutations are tallied as the baseline mutation rate against which the potential hypermutations must be compared.
A one-sided Fisher’s exact test is used to compare the proportion of GRD positions that became ARD positions to the proportion of GY or GRC positions that became either AY or ARC positions. When the p-value of the test is smaller than some threshold, with the default set to 0.1 as in (Abrahams et al., 2009), then the individual sequence is flagged as a hypermutant and either the sequence is removed from the dataset, or the mutated bases (the A’s followed by RD) are replaced by an R to indicate that we are uncertain whether the mutation was a random mutation or if it was the result of hypermutation.
In order to be a position of interest (either a control or hypermutation position), all that is required is a G in the ancestral sequence. To classify the position into either a hypermutation or control position, only the query sequence is considered. If the two positions following the position that contains the G in the ancestral sequence matches RD, then it is a hypermutation position, else it is a control position. The two downstream positions in the ancestral sequence are not considered. This implies the assumption that the two downstream positions in the ancestral sequence mutates before the position of interest.
Make sure you have a recent version of R. Follow the instructions in the following link to set up the correct repositiory for apt: http://stackoverflow.com/questions/10476713/how-to-upgrade-r-in-ubuntu.
Make sure that both r-base and r-base-dev is installed
sudo apt-get install r-base r-base-dev
Next, install devtools' depedancies with apt-get:
sudo apt-get install libssl-dev libxml2-dev libcurl4-gnutls-dev
Then, from within R, install devtools:
install.packages('devtools', repo = 'http://cran.rstudio.com/')
Install hypermutR from github:
library(devtools)
install_github('philliplab/hypermutR')
Lastly, hypermutR includes a script that can be run from the commandline. You need to put this script somewhere convenient ('/usr/bin' for example)
file.symlink(from = file.path(find.package('hypermutR'), 'hypermutR.R'),
to = '/usr/bin')
library(hypermutR)
help('remove_hypermutation')
This will display the help for the main function in hypermutR.
hypermutR -h
or (depending on your installation):
hypermutR.R -h
This will display help for all the options and an example call to hypermutR.
More details can be found in the vignette, which can be viewed online at hypermutR vignette