The script mag_refinement.sh is for extending contigs obtained from a metagenome assembly. Its primary purpose lies in improving size and contiguity of contigs representing a metagenome assembled genome (MAG). Once a MAG is obtained (either by manual binning of contigs or by using any available binning software), this script can be used to elongate, and potentially connect, the contigs of the MAG. This is done by mapping of all reads used in the initial metagenome assembly to the contigs of the MAG, and then reassembling the mapped reads. This process is then repeated several times, always using the contigs from the previous iteration (only using those >2kb) as input for the mapping in the current iteration.

mag_refinement.sh currently also uses two helper scripts, fasta_to_gc_cov_length_tab.pl and tab_to_fasta.pl, to filter the assembled contigs by length. I recommend adding the path to those two helper scripts to your system's path or to your .bashrc file (alternatively you could add the absolute path to the two scripts into mag_refinement.sh).

The current version of the script was developed to include short-read (Illumina paired end) as well as long-read (Pacbio HIFI reads) metagenome data. The mapped long reads are included in the assembly as single end reads. However, the script could easily be modified to run on short-read data only by removing the lines related to the mapping of the pacbio reads and removing the argument for the single end reads in the assembly.

mag_refinement.sh uses the following publically available software (with the respective version I am currently using the script with in brackets): minimap2 (v2.22-r1101), samtools (v1.14), coverM (v0.6.1), SeqKit (v2.3.0), SPAdes (v3.15.3).

For mag_refinement.sh to successfully find all the dependencies, minimap2, samtools, coverM and SeqKit need to be installed in a conda environment called "read_mapping", while SPAdes needs to be installed in a conda environment called "assembly".

mag_refinement.sh has been developed on Linux 4.9.0-17-amd64 x86_64 using Debian 9

Just download mag_refinment.sh plus the two helper scritps (fasta_to_gc_cov_length_tab.pl and tab_to_fasta.pl) into your desired directory.

./mag_refinment.sh seed_contigs_path pacbio_path fw_reads_path rv_reads_path project_name no_of_iterations threads
with
seed_contigs_path path to fasta file with seed contigs
pacbio_path path to pacbio reads
fw_reads_path path to forward reads
rv_reads_path path to reverse reads
project_name prefix for output files
no_of_iterations number of iterations to run
threads number of threads to use where appropriate
While the script is running, you can check the number of contigs and the (rough) total size of each iteration with the two commands wc -c it*/spades*/*filtered.fa and wc -l it*/spades*/*filtered.tab. Once the total size and number of contigs is stable, you might terminate the script and start looking at your assemblies. I highly recommend checking (and if necessary refining) the assembled contigs from iterations of interest using the anvi'o programm anvi-refine.

To test the script, we are providing a artificial metagenome reads together with an artificial MAG. The files can be downloaded from here: test dataset.
The artificial metagenome represents artificially created reads (illumina paired end (150bp) plus pacbio (8kb)) for five marine bacteria (Candidatus Pelagibacter ubique HTCC1062 (GCA_000012345.1), Maribacter sp. HTCC2170 (GCA_000153165.2), Polaribacter sp. MED152 (GCA_000152945.2), Hyphomonas neptunium ATCC 15444 (GCA_000013025.1) and Synechococcus sp. CC9311 (GCA_000014585.1)).
The artificial reads were created using randomreads.sh (part of bbtools).
An artificial MAG was created for Candidatus Pelagibacter ubique HTCC1062, comprising nine 100kb fragments with 50kb gaps in between (except for the gap between the last and the first contig, which is only ~8kb).

Running mag_refinement.sh with our test dataset using 50 threads and 30 iterations (~42 min runtime), the size of the MAG increased from 900kb to 1.2Mb, while the number of contigs decreased from 9 to 7, thus improving size and continguity of the MAG (it is likely that the MAG could be improved even further by running more iterations).