metaVaR: MetaVariant Species (MVS) for reference-free and metagenomic-based population genomic analysis.

Why using metaVaR?

metaVaR is a R package developed for population genomic analyses based on variants generated directly from metagenomic data without reference genomes or transcriptomes. Its utilization is based on the output of the reference-free variant caller DiscoSnp++ ran on multisamples metagenomic reads. metaVaR clusters the metavariants into species called metavariant species (MVS). metaVaR allows to estimate genomic differentiation through F-statistics and to identify loci under natural selection.

How to install metaVaR?

metaVaR is a R package that can be installed directly from github as follow:

install_github("madoui/metaVaR")

The package has been submitted to CRAN and will be available from it soon via install.packages().

How to use metaVaR?

Quick running example on a toyset

One toysets can be used with the package, the "MS5" dataset, created from DiscoSNP++ ran on real metagenomic data collected in five sampling sites in the Mediterranean Sea. Here is an example to run metaVaR on the "MS5" dataset.

library(metaVaR)

# espilon values to test for dbscan
e = c(3,5,7)
# minimum points values to test for dbscan
p = c(5, 10, 20, 50)

# Test all dbscan parameter couples
MVC = tryParam (e, p, MS5$cov)

# Select maximum weigthed independant sets
MWIS = getMWIS (MVC)

# Filter loci to generate MVS
MVS = mvc2mvs(MWIS, freq = MS5$freq)

The results consists in a list of objects of class mvs that contains the allele frequency matrix, global and pairwise $F_{ST}$, LK and associated corrected p-value.

Data preprocessing and input data format

You first need to run DiscoSnp++ on multisample metagenomic data without references. Then, use the DiscoSnp++ VCF output to generate two matices: one depth of coverage matrix for bialleic loci and one allele frequency matrix. This task can be performed using the perl script metavarFilter.pl designed to parse the DiscoSnp++ output.

metaVarFilter.pl -i input.vcf -p output_prefix [-a minCov -b maxCov -c minPop]         
	Input:
	-i	a VCF file produced by DiscoSNP++ ran on multisample metagenomic data
	-p	a prefix output
  
  Options:
  -a      The minimum cumulative depth of coverage of a loci across populations (default 20)
  -b      The maximum cumulative depth of coverage of a loci across populations (default 250)
	-c	    The minimum number of populations avec a SNP call for a loci to be kept (default 4)
	
	Outputs:
	1. prefix_filtered.vcf		a filtered VCF file that contains informative biallelic SNVs
	2. prefix_cov.txt		      a file with the depth of coverage of biallelic loci in each population
	3. prefix_freq.txt		    a file with the allele frequencies of biallelic loci in each population

-a and -b parameters can be set to 0 and 10000 if all snps want to be kept, but this will slow the clustering. To get rid of noisy snps, the distribibution of the cumulative depth of coverage can be observed. This will show a bell shape curve, -a and -b can be the oberved lower and upper limits of the bell.

Read/Write metaVaR outputs

# run metaVaR
e = c(3,5)
p = c(5,10)
MVC_list = tryParam (e, p, MS5$cov)

# store MVC
writeMvcList (MVC_list, "MVC")

# import MVC
MVC_list = readMvcList ("MVC")

MWIS_list = getMWIS (MVC_list)
MVS_list = mvc2mvs(MWIS_list, freq = MS5$freq)

# store MVS
writeMvsList (MVS_list, "MVS")

# read MVS
MVS_list = readMvsList ("MVS")

Running metaVaR on HPC

To take advantages of the metaVaR custering algorithm (mDBSCAN-WMIN), a lot of epsilon and minimum points values can be tested on grid computing. In this case, all MVC_list have to be written in the same output directory. Then, the mvc2mvs can be executed on all MVC.

Results visualisation

There are currently four types of plots that ca be produced on a MVS object. Here an example on the first mvs of the list 'MVS'.

# Plot the MVS distribution of allele frequency
plotMvs (MVS[[1]], type = "freq")

# Plot the MVS loci depth of coverage
plotMvs (MVS[[1]], type = "cov")

# Plot the pairwise $F_{ST}$ matrix
plotMvs (MVS[[1]], type = "heatFst")

# Plot the expected and observed LK distribution
plotMvs (MVS[[1]], type = "LK")

How to cite metaVaR?

metaVaR: introducing metavariant species models for reference-free metagenomic-based population genomics. Romuald Laso-Jadart, Christophe Ambroise, Pierre Peterlongo, Mohammed-Amin Madoui. (2020) bioRxiv doi: https://doi.org/10.1101/2020.01.30.924381

pascalhabluetzel / metaVaR