MAGinator

Combining the strengths of contig and gene based methods to provide:

• Accurate abundances of species using de novo signature genes
  • MAGinator uses a statistical model to find the best genes for calculating accurate abundances
• SNV-level resolution phylogenetic trees based on signature genes
  • MAGinator creates a phylogenetic tree for each species so you can associate your metadata with subspecies/strain level differences
• Connect accessory genome to the species annotation by getting a taxonomic scope for gene clusters
  • MAGinator clusters all ORFs into gene clusters and for each gene cluster you will know which taxonomic level it is specific to
• Improve your functional annotation by grouping your genes in synteny clusters based on genomic adjacency
  • MAGinator clusters gene clusters into synteny clusters - Syntenic genes are usually part of the same pathway or have similar functions

Installation

All you need for running MAGinator is snakemake and mamba. Other dependencies will be installed by snakemake automatically.

conda create -n maginator -c bioconda -c conda-forge snakemake mamba
conda activate maginator
pip install maginator

Furthermore, MAGinator also needs the GTDB-tk database downloaded. Here we download release 214. If you don't already have it, you can run the following:

wget https://data.ace.uq.edu.au/public/gtdb/data/releases/release214/214.1/auxillary_files/gtdbtk_r214_data.tar.gz
tar xvzf *.tar.gz

Usage

MAGinator needs 3 input files:

• The clusters.tsv files from VAMB
• A fasta file with sequences of all contigs, with unique names
• A comma-separated file giving the position of the fastq files with your sequencing reads formatted as: SampleName,PathToForwardReads,PathToReverseReads

Run MAGinator:

maginator -v vamb_clusters.tsv -r reads.csv -c contigs.fasta -o my_output -g "/path/to/GTDB-Tk/database/release214/"

A testset can be found in the test_data directory.

1. Download the 3 samples used for the test at SRA: https://www.ncbi.nlm.nih.gov/sra?LinkName=bioproject_sra_all&from_uid=715601 with the ID's dfc99c_A, f9d84e_A and 221641_A
2. Change the paths to the read-files in reads.csv
3. Unzip the contigs.fasta.gz
4. Run MAGinator

Run on a compute cluster

MAGinator can run on compute clusters using qsub (torque), sbatch (slurm), or drmaa structures. The --cluster argument toggles the type of compute cluster infrastructure. The --cluster_info argument toggles the information given to the submission command, and it has to contain the following keywords {cores}, {memory}, {runtime}, which are used to forward resource information to the cluster.

A qsub MAGinator can for example be run with the following command (... indicates required arguments, see above):

maginator ... --cluster qsub --cluster_info "-l nodes=1:ppn={cores}:thinnode,mem={memory}gb,walltime={runtime}"

Test data

A test set can be found in the maginator/test_data directory.

1. Download the 3 samples used for the test at SRA: https://www.ncbi.nlm.nih.gov/sra?LinkName=bioproject_sra_all&from_uid=715601 with the ID's dfc99c_A, f9d84e_A and 221641_A
2. Clone repo: git clone https://github.com/Russel88/MAGinator.git
3. Change the paths to the read-files in reads.csv
4. Unzip the contigs.fasta.gz
5. Run MAGinator

MAGinator can been run on the test data on a slurm server with the following command:

maginator --vamb_clusters clusters.tsv --reads reads.csv --contigs contigs.fasta --gtdb_db data/release214/ --output test_out --cluster slurm --cluster_info "-n {cores} --mem {mem_gb}gb -t {runtime}" --max_mem 180

The expected output can be found as a zipped file on Zenodo: https://doi.org/10.5281/zenodo.8279036. MAGinator has been run on the test data (using GTDB-tk db release207_v2) on a slurm server.

On the compute cluster each job have had access to 180gb RAM, with the following time consumption: real 72m27.379s user 0m18.830s sys 1m0.454s

If you run on a smaller server you can set the parameters --max_cores and --max_mem.

Recommended workflow

To generate the input files to run MAGinator we have created a recommended workflow, with preprocessing, assembly and binning* of your metagenomics reads (the rules for binning have been copied from VAMB (https://github.com/RasmussenLab/vamb/blob/master/workflow/)). It has been setup as a snakefile in recommended_workflow/reads_to_bins.Snakefile.

The input to the workflow is the reads.csv file. The workflow can be run using snakemake:

snakemake --use-conda -s reads_to_bins.Snakefile --resources mem_gb=180 --config reads=reads.csv --cores 10 --printshellcmds 

Once the binning is done, we recommend using a tool like dRep (https://github.com/MrOlm/drep) to create the species-level clusters. The advantage of dRep is that the clustering parameters can be modified to create clusters that belong to different taxonomic levels. An R script is located in recommended workflow/MAGinator_setup.R, that will create the different input files for MAGinator adapted to the output of dRep.

Preparing data for MAGinator run

sed 's/@/_/g' assembly/all_assemblies.fasta > all_assemblies.fasta
sed 's/@/_/g' vamb/clusters.tsv > clusters.tsv

Now you are ready to run MAGinator.

Functional Annotation

To generate the functional annotation of the genes we recommend using EggNOG mapper (https://github.com/eggnogdb/eggnog-mapper).

You can download it and try to run it on the test data

mkdir test_out/functional_annotation
emapper.py -i test/genes/all_genes_rep_seq.fasta --output test_out/functional_annotation -m diamond --cpu 38

The eggNOG output can be merged with clusters.tsv and further processed to obtain functional annotations of the MAG, cluster or sample levels with the following command:

(echo -e '#sample\tMAG_cluster\tMAG\tfunction'; join -1 1 -2 1 <(awk '{print $2 "\t" $1}' clusters.tsv | sort) <(tail -n +6 annotations.tsv | head -n -3 | cut -f1,15 | grep -v '\-$' | sed 's/_[[:digit:]]\+\t/\t/' | sed 's/,/\n/g' | perl -lane '{$q = $F[0] if $#F > 0; unshift(@F, $q) if $#F == 0}; print "$F[0]\t$F[1]"' | sed 's/\tko:/\t/' | sort) | awk '{print $2 "\t" $2 "\t" $3}' | sed 's/_/\t/' | sort -k1,1 -k2,2n) > MAGfunctions.tsv

In this case the KEGG ortholog column 15 was picked from the eggNOG-mapper output. But by cutting e.g. column number 13, one would obtain GO terms instead. Refer to the header of the eggNOG-mapper output for other available functional annotations e.g. KEGG pathways, Pfam, CAZy, COGs, etc.

MAGinator workflow

This is what MAGinator does with your input (if you want to see all parameters run maginator --help):

• Filter bins by size
  • Use --binsize to control the cutoff
• Run GTDB-tk to taxonomically annotate bins and call open reading frames (ORFs)
• Group your VAMB clusters into metagenomic species (MGS) based on the taxonomic annotation. (Unannotated VAMB clusters are kept in the pipeline, but left unchanged)
  • Use --no_mgs to disable this
  • Use --annotation_prevalence to change how prevalent an annotation has to be in a VAMB cluster to call taxonomic consensus
• Cluster your ORFs into gene clusters to get a non-redundant gene catalogue
  • Use --clustering_min_seq_id to toggle the clustering identity
  • Use --clustering_coverage to toggle the clustering coverage
  • Use --clustering_type to toggle whether to cluster on amino acid or nucleotide level
• Map reads to the non-redundant gene catalogue
  • Use --min_length to filter for the minimum number of basepairs that must be aligned to keep a read
  • Use --min_identity to filter for the minimum percentage of identity of mapped read to be kept
  • Use --min_map to filter for the minimum percentage of a read that has to be mapped to be kept
• Create a gene count matrix based on a signature reads approach
  • By default, MAGinator will redistribute ambiguous mapping reads based on the profile of uniquely mapping reads
  • This can be changed with the --multi option.
• Pick non-redundant genes that are only found in one MAG cluster each
• Fit signature gene model and use the resulting signature genes to get the abundance of each MAG cluster
  • Use --min_mapped_signature_genes to change minimum number of signature genes to be detected in the sample to be included in the analysis
  • Use --min_samples to alter the number of samples with the MAG cluster present in order to perform signature gene refinement
• Prepare for generation of phylogenies for each MAG cluster by finding outgroups and marker genes which will be used for rooting the phylogenies
• Use the read mappings to collect SNV information for each signature gene and marker gene for each sample
• Align signature and marker genes, concatenate alignments and infer phylogenetic trees for each MAG cluster
  • Use --phylo to toggle whether use fasttree (fast, approximate) or iqtree (slow, precise) to infer phylogenies
• Infer the taxonomic scope of each gene cluster. That is, at what taxonomic level are genes from a given gene cluster found in
  • Use --tax_scope_threshold to toggle the threshold for how to find the taxonomic scope consensus
• Cluster gene clusters into synteny clusters based on how often they are found adjacent on contigs

Output

• abundance/
  • abundance_phyloseq.RData - Phyloseq object for R, with absolute abundance and taxonomic data
• clusters/
  • /.fa - Fasta files with nucleotide sequence of bins
• genes/
  • all_genes.faa - Amino acid sequences of all ORFs
  • all_genes.fna - Nucletotide sequences of all ORFs
  • all_genes_nonredundant.fasta - Nucleotide sequences of gene cluster representatives
  • all_genes_cluster.tsv - Gene clusters
  • matrix/
    • gene_count_matrix.tsv - Read count for each gene cluster for each sample
    • small_gene_count_matrix.tsv - Read count matrix only containing the genes, that does not cluster across MAG cluster
  • synteny/ - Intermediate files for synteny clustering of gene clusters
• gtdbtk/
  • / - GTDB-tk taxonomic annotation for each VAMB cluster
• logs/ - Log files
• mapped_reads/
  • bams/ - Bam files for mapping reads to gene clusters
• phylo/
  • alignments/ - Alignments for each signature gene
  • cluster_alignments/ - Concatenated alignments for each MAG cluster
  • pileup/ - SNV information for each MAG cluster and each sample
  • trees/ - Phylogenetic trees for each MAG cluster
  • stats.tab - Mapping information such as non-N fraction, number of signature genes and marker genes, read depth, and number of bases not reaching allele frequency cutoff
  • stats_genes.tab - Same as above but the information is split per gene
• signature_genes/
  • - R data files with signature gene optimization
  • read-count_detected-genes.pdf - Figure for each MAG cluster displaying number of identified SG's in each sample along with the number of reads mapped.
• signature_reads/
  • profiles - Read count profiles with ambiguous reads redistributed based on the uniquely mapped reads profile
• tabs/
  • gene_cluster_bins.tab - Table listing which bins each gene cluster was found in
  • gene_cluster_tax_scope.tab - Table listing the taxonomic scope of each gene cluster
  • metagenomicspecies.tab - Table listing which, if any, clusters where merged in MAG cluster and the taxonomy of those
  • signature_genes_cluster.tsv - Table with the signature genes for each MAG cluster
  • synteny_clusters.tab - Table listing the synteny cluster association for the gene clusters. Gene clusters from the same synteny cluster are genomically adjacent.
  • tax_matrix.tsv - Table with taxonomy information for MAG cluster