Skip to content

Dereplication, and Coverage Inspection of MAGs

In this final part of the tutorial we aggregate the MAGs recovered from multiple samples, dereplicate them, and briefly inspect both the resulting clusters and the read-mapping depth (coverage) of some MAGs in the small example dataset.

Dereplication reduces a large set of MAGs to a non-redundant genome catalogue by grouping nearly identical genomes (typically using Average Nucleotide Identity, ANI) and selecting the highest-quality MAG as the representative for each cluster. This avoids double-counting strains, simplifies downstream analyses and prevents ambiguous read mapping to many almost-identical genomes.

Within the Metagenomics-Toolkit, dereplication is implemented as a separate module that:

  1. takes the previous run or a table describing all MAGs (including completeness, contamination and coverage),
  2. clusters similar genomes in a bottom-up fashion (using Mash and ANI distances), and
  3. selects the best representative per cluster based on quality and assembly statistics.

Metagenomics-Toolkit Execution

The following lines represent the part of the configuration that tells the Toolkit to run the Dereplication module on your existing results.

Dereplication Configuration File Snippet
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
output: "output"
input: "output"
logDir: log
runid: 1
logLevel: 1
scratch: false
publishDirMode: "copy"
steps:
  dereplication:
    bottomUpClustering:
      # stricter MIMAG medium quality
      minimumCompleteness: 50
      maximumContamination: 5
      ANIBuffer: 20
      mashBuffer: 2000
      method: 'ANI'
      additionalParams:
        mash_sketch: ""
        mash_dist: ""
        # cluster cutoff
        cluster: " -c 0.05 "
        pyani: " -m ANIb "
        representativeAniCutoff: 0.95
  readMapping:
    bwa2:
      additionalParams:
        bwa2_index: ""
        bwa2_mem: ""
     # This module produces two abundance tables.
     # One table is based on relative abundance and the second one on the trimmed mean.
     # Just using relative abundance makes it difficult to tell if a genome is part of a dataset.
     # Thats why it makes sense to set at leat a low min covered fraction parameter.
    coverm: " --min-covered-fraction 80  --min-read-percent-identity 95 --min-read-aligned-percent 95 "
    covermONT: " --min-covered-fraction 80  --min-read-aligned-percent 95 "
    minimap:
      additionalParams:
        minimap_index: ""
        minimap: ""

For reference, your complete parameter file for this final step looks like this:

Parameter-file 
tempdir: "tmp"
summary: false
s3SignIn: false
output: "output"
input: "output"
logDir: log
runid: 1
logLevel: 1
scratch: false
publishDirMode: "copy"
steps:
  dereplication:
    bottomUpClustering:
      # stricter MIMAG medium quality
      minimumCompleteness: 50
      maximumContamination: 5
      ANIBuffer: 20
      mashBuffer: 2000
      method: 'ANI'
      additionalParams:
        mash_sketch: ""
        mash_dist: ""
        # cluster cutoff
        cluster: " -c 0.05 "
        pyani: " -m ANIb "
        representativeAniCutoff: 0.95
  readMapping:
    bwa2:
      additionalParams:
        bwa2_index: ""
        bwa2_mem: ""
     # This module produces two abundance tables.
     # One table is based on relative abundance and the second one on the trimmed mean.
     # Just using relative abundance makes it difficult to tell if a genome is part of a dataset.
     # Thats why it makes sense to set at leat a low min covered fraction parameter.
    coverm: " --min-covered-fraction 80  --min-read-percent-identity 95 --min-read-aligned-percent 95 "
    covermONT: " --min-covered-fraction 80  --min-read-aligned-percent 95 "
    minimap:
      additionalParams:
        minimap_index: ""
        minimap: ""
resources:
  highmemLarge:
    cpus: 28
    memory: 230
  highmemMedium:
    cpus: 14
    memory: 113
  large:
    cpus: 28
    memory: 58
  medium:
    cpus: 14
    memory: 29
  small:
    cpus: 7
    memory: 14
  tiny:
    cpus: 1
    memory: 1

We assume that all required steps by the previous tutorials are done.

If you run the Toolkit again with the -resume flag and the above dereplication configuration, it will reuse all previous results (QC, assembly, binning, MagAttributes) and only execute the new dereplication step.

Task 1

Change into the directory of the Toolkit session. 

cd /vol/volume/sessions/metagenomics_metagenomics-tk

Run the dereplication module on your results using the following command: 

NXF_VER=25.04.2 nextflow run metagenomics/metagenomics-tk \
      -profile standard \
      -params-file https://raw.githubusercontent.com/metagenomics/metagenomics-tk/refs/heads/doc/exhibition-day/tutorial/default/tutorials/tutorial2/fullPipeline_dereplication.yml \
      -ansi-log false \
      -entry wAggregatePipeline \
      -resume \
    --input output \
      --logDir dereplication

This will:

  • collect all the necessary results from your MAGs to create the summary table needed as input,
  • cluster similar genomes based on ANI, and
  • write a clusters.tsv file describing the dereplicated genome clusters.

Output

Dereplication Clusters

The example dereplication input table looks like this (one MAG per row):

BIN_ID  COMPLETENESS    COVERAGE        CONTAMINATION   HETEROGENEITY   PATH    N50
SRR492065_bin.3.fa      36.29   171.259 0.39    0       metagenomics-tk/work/58/267ffbad749945cb7dfd98b23a7449/SRR492065_bin.3.fa      73239
SRR492183_bin.2.fa      100.0   142.023 0.1     0       metagenomics-tk/work/90/b5b80ef93e064e7d723eca9bd92241/SRR492183_bin.2.fa      175163
SRR492065_bin.1.fa      85.99   14.7729 18.95   0       metagenomics-tk/work/58/267ffbad749945cb7dfd98b23a7449/SRR492065_bin.1.fa      7308
SRR492065_bin.5.fa      64.75   152.837 0.54    0       metagenomics-tk/work/58/267ffbad749945cb7dfd98b23a7449/SRR492065_bin.5.fa      161674
SRR492065_bin.4.fa      96.32   70.5657 0.14    0       metagenomics-tk/work/58/267ffbad749945cb7dfd98b23a7449/SRR492065_bin.4.fa      25145
SRR492065_bin.2.fa      94.64   23.2949 0.11    0       metagenomics-tk/work/58/267ffbad749945cb7dfd98b23a7449/SRR492065_bin.2.fa      16802
SRR492183_bin.3.fa      92.84   27.7866 0.08    0       metagenomics-tk/work/90/b5b80ef93e064e7d723eca9bd92241/SRR492183_bin.3.fa      21121
SRR492183_bin.1.fa      99.96   45.5264 0.01    0       metagenomics-tk/work/90/b5b80ef93e064e7d723eca9bd92241/SRR492183_bin.1.fa      46795

The dereplication module uses these columns as follows:

  • COMPLETENESS, CONTAMINATION – to filter MAGs before clustering.
  • COVERAGE, N50, HETEROGENEITY – to decide which genome becomes the representative of each cluster.

After running Task 1, dereplication produces a clusters.tsv file with one row per MAG and three columns:

  • CLUSTER – cluster ID (all genomes in the same cluster are near-identical),
  • GENOME – path or identifier of the genome,
  • REPRESENTATIVE – 1 if the genome is the chosen representative, 0 otherwise.

A typical file will look like this:

CLUSTER  GENOME                                 REPRESENTATIVE
1        bin.1.fa                1
1        bin.2.fa                0
2        bin.8.fasta             1
2        bin.9.fasta             0
3        bin.10.fasta            0
3        bin.32.fa               1

(The exact clustering in your run may differ; this is just an illustrative example.)

Task 2

Locate the dereplication clusters file and inspect how many clusters were formed and which genomes were chosen as representatives. Hint: There is a new directory in your output folder.

Solution

Change into the Toolkit session directory (if you are not already there):

cd /vol/volume/sessions/metagenomics_metagenomics-tk

First, locate the new directory in your output folder:

ls -la output/
There is a new AGGREGATED directory, lets look at our clusters.tsv in the similarly named directory.

cat output/AGGREGATED/1/dereplication/0.1.1/bottomUpClustering/clusters/clusters.tsv

It should look similar to this:

CLUSTER GENOME  REPRESENTATIVE
1       SRR492065_bin.2.fa      0.0
3       SRR492065_bin.4.fa      1.0
2       SRR492065_bin.5.fa      0.0
2       SRR492183_bin.1.fa      1.0
4       SRR492183_bin.2.fa      1.0
1       SRR492183_bin.3.fa      1.0

This shows you: * how many clusters exist (number of unique values in the CLUSTER column), and
* which MAG(s) per cluster have REPRESENTATIVE=1. Two MAGs where redundant and are now part of clusters represented only by its best MAG.

Read Mapping Depth (Coverage) of Example MAGs

Coverage values are derived from mapping reads from all input samples back to each representative MAG.

Task 3

Let's have a look to examine the coverage (read mapping depth) of our representative MAGs. Explore the new directory, find the readMapping folder and look for the relative abundance coverage of our representatives with all SRR492065 reads.

Solution

Change again into the Toolkit session directory:

cd /vol/volume/sessions/metagenomics_metagenomics-tk

Explore the AGGREGATED directory:

ls -la output/AGGREGATED/1/

There is a readMapping directory. If we explore further:

ls -la output/AGGREGATED/1/readMapping/0.1.0/
Also a genomeCoverage one. If we look at this, we see something like:

-rw-r--r-- 1 ubuntu ubuntu  111 Dec  9 18:50 SRR492065_count.tsv
-rw-r--r-- 1 ubuntu ubuntu  120 Dec  9 18:50 SRR492065_mean.tsv
-rw-r--r-- 1 ubuntu ubuntu  138 Dec  9 18:50 SRR492065_relative_abundance.tsv
-rw-r--r-- 1 ubuntu ubuntu  120 Dec  9 18:50 SRR492065_rpkm.tsv
-rw-r--r-- 1 ubuntu ubuntu  116 Dec  9 18:50 SRR492065_tpm.tsv
-rw-r--r-- 1 ubuntu ubuntu  118 Dec  9 18:50 SRR492065_trimmed_mean.tsv
-rw-r--r-- 1 ubuntu ubuntu  110 Dec  9 18:50 SRR492183_count.tsv
-rw-r--r-- 1 ubuntu ubuntu  113 Dec  9 18:50 SRR492183_mean.tsv
-rw-r--r-- 1 ubuntu ubuntu  130 Dec  9 18:50 SRR492183_relative_abundance.tsv
-rw-r--r-- 1 ubuntu ubuntu  110 Dec  9 18:50 SRR492183_rpkm.tsv
-rw-r--r-- 1 ubuntu ubuntu  112 Dec  9 18:50 SRR492183_tpm.tsv
-rw-r--r-- 1 ubuntu ubuntu  113 Dec  9 18:50 SRR492183_trimmed_mean.tsv
drwxr-xr-x 2 ubuntu ubuntu 4096 Dec  9 18:50 logs
In SRR492065_relative_abundance.tsv we find the right abundances.

cat output/AGGREGATED/1/readMapping/0.1.0/genomeCoverage/SRR492065_relative_abundance.tsv

Genome  SRR492065
unmapped        21.920425
SRR492065_bin.4 22.608734
SRR492183_bin.1 45.45554
SRR492183_bin.2 2.6764407
SRR492183_bin.3 7.338865