

MAGs from binning metagenomes

What is a MAG?

A MAG (metagenome-assembled genome) is a putative genome, reconstructed from metagenomic data. Because these genomes do not derive from a clonal culture they typically represent a consensus of the genomes of many closely related cells in one or more metagenomic samples, and for this reason they are sometimes called population genomes. It is important to remember that, because of the way MAGs are inferred from the data, the resulting sequence is usually fragmentary and may not accurately represent the genome of any cell in the community. It is merely an average estimate, and there are well-known problems with taking averages, see for example the wikipedia page on Simpson's paradox to get an idea of how averages can go wrong.

What is a good MAG?

The international genomics community has made an effort to define quality standards for MAGs via the Minimum Information about a Metagenome Assembled Genome (MIMAG) standards. The basic idea is to require that a MAG possess a certain minimum level of estimated completeness and maximal level of estimated contamination for it to be considered as one of "Low-quality", "Medium-quality", "High-quality", or "Finished". See Table 1 in the above-linked paper for the full details. Generally speaking it is hard to achieve "High-quality" and nearly impossible to get "Finished" with short read sequencing data alone. Long read metagenome data has been shown to produce results in these categories, although it can be difficult to obtain sufficient DNA for those methods, and they currently remain more expensive than short read sequencing.

Making MAGs

The process of reconstructing genomes from a metagenome is often referred to as metagenome binning or just binning, from the process of assigning contigs to one 'bin' per genome. There are many binning tools available to extract MAGs from metagenome assemblies. When timeseries data is available, MetaBAT2 is a good choice because it is both easy to use and offers good performance. MetaBAT2 can be installed via conda as follows:

Install MetaBAT2

conda install -y -n workshop metabat2

MetaBAT2 input data

As input, MetaBAT2 requires reads from each of the shotgun metagenome samples to be mapped back to the metagenome assembly. We will not cover the read mapping process in this tutorial, but it can be carried out using standard mapping software such as bwa mem or bowtie2. Be sure to sort and index the bam files with samtools prior to running metabat.

Assuming we have bam files of mapped reads and the metagenome assembly available in a directory called /data/assembly we can compute genome bins as follows:

Analyse the assembled timeseries data-set with runMetaBat.sh

# create a folder for metabat output
mkdir ~/metabat ; cd ~/metabat
# analyze the set of bams against the metagenome
runMetaBat.sh /data/assembly/contigs-fixnames.fa /data/assembly/*.bam 

Depending on how much data you've got this can take a long time to compute. Luckily MetaBAT2 has a good parallel implementation so it can go faster if you run on a large multi-core machine. At the end of the process MetaBAT2 will produce a directory called contigs-fixnames.fa.metabat-bins, which contains one FastA file of contigs for each genome bin that it inferred.