Using FreeBayes and deepSNV to call variants in mixed populations

The program FreeBayes can be used to call variants in genomes of any ploidy, pooled samples, or mixed populations.

Installation for FreeBayes

This tutorial assumes that you have created the paths $WORK/src and $HOME/local/bin and added $HOME/local/bin to your $PATH. FreeBayes uses a git repository and requires the cmake build system to compile. You can install it with these commands:

login1$ module load git
login1$ mkdir -p $WORK/src && cd $WORK/src
login1$ git clone --recursive git://github.com/ekg/freebayes.git
login1$ cd freebayes
login1$ module load cmake
login1$ module load gcc
login1$ make
login1$ mv bin/* $HOME/local/bin

This command from the instructions attempts to install to a system-wide location as super-user:

sudo make install

This won't work on Lonestar! (You aren't an admin.) However, the make command created the executables inside of the source tree, so we find and move them to our standard $HOME/local/bin directory with the last command.

Installation for deepSNV

The newest R module on lonestar is version 2.14, but deep SNV requires R 2.15.

You can install your own version of R 2.15 using the instructions below, but this takes a while to compile, so you can also just add this location to your path by adding this line to your ~/.profile_user file.

export PATH="/corral-repl/utexas/BioITeam/ngs_course/local/bin:$PATH"

If you want to go through installing R 2.15 and deep SNV for yourself, here's how:

login1$ wget http://cran.wustl.edu/src/base/R-2/R-2.15.0.tar.gz
login1$ tar -xvzf R-2.15.0.tar.gz
login1$ cd R-2.15.0
login1$ ./configure --prefix=$HOME/local
login1$ make
login1$ make install

Once you have access to R 2.15, you can install deepSNV using these commands (which work for anyBioConductor package.

login1$ R
...
> source("http://bioconductor.org/biocLite.R")
> biocLite("deepSNV")

Example 1: Mixed E. coli population

A mixed population of E. coli from an evolution experiment was sequenced at several different time points (PMID: 19838166 , PMID:19776167). At generation 0 it consisted of a clone (cells grown from a colony with essentially no genetic variation), then additional samples were taken at 20K and 40K generations after which mutations arose and swept through the (asexual) population.

Data

The data files for this example are in the path:

/corral-repl/utexas/BioITeam/ngs_course/ecoli_mixed

The read files were downloaded from the ENA SRA study.

So that we can treat all the data as single-ended for simplicity, we concatenated two separate FASTQ (paired-end) files for sample SRR030252 using this command

cat SRR030252_1.fastq SRR030252_2.fastq > SRR030252.fastq

The reference genome file was downloaded from the NCBI Genomes page.

We renamed the FASTA sequence from "gi|254160123|ref|NC_012967.1|" to "NC_012967" by changing the first line of the NC_012967.1.fasta sequence using a text editor.

Map Reads

Choose an appropriate program and map the reads. As for other variant callers, convert the mapped reads to BAM format, then sort and index the BAM file.

Additional exercises

• Determine the approximate depth of mapped read coverage for each sequencing data set.
• Try using different mappers or changing the default alignment settings to find more variants.

Run FreeBayes

FreeBayes can be used to treat the sample as a mixture of pooled samples. (In our case it is actually a mixture of >1 million bacteria, but we have nowhere near that level of coverage, so we give an arbitrary mixed ploidy of 100, which means we use a statistical model that predicts variants only with frequencies of 1%, 2%, 3%, ... 98%, 99%, 100%). This command runs pretty fast, so you can do it in interactive mode.

login1$ freebayes --min-alternate-total 2 --ploidy 100 --pooled --vcf SRR032374.vcf \
        --fasta-reference NC_012967.1.fasta SRR032374.sorted.bam

Additional exercises

• Write a script or use a linux command to filter the output files to only contain variants that are predicted to have frequencies > 0.05 or scores > 1000.

Run deepSNV

deepSNV runs more slowly, so we will only look at a small region of the genome initially in interactive mode. (Why is it slower? Probably in part due to using a more sophisticated statistical model and in part because it is implemented in R instead of C.)

Useful Links

• deepSNV website
• deepSNV paper .. but UT library does not have access to this journal...
• News article about deepSNV
• deepSNV R module vignette

login1$ R
...
> regions <- data.frame(chr="NC_012967", start = 1, stop=100000)
> result = deepSNV(test = "SRR032376.sorted.bam", control = "SRR030252.sorted.bam", regions=regions)
> sig_result <- summary(result, sig.level=0.05, adjust.method="BH")
> pdf("SRR032374.pdf")
> plot(mix)
> dev.off()
> write.csv(sig_result, "SRR032374")

Additional exercises

• Create an R script to run from the command line to execute these commands, and try running the entire E. coli genome on TACC.
• Compare the variants predicted in samples SRR032374 and SRR032376.