dadasnake is a Snakemake workflow to process amplicon sequencing data, from raw fastq-files to taxonomically assigned "OTU" tables, based on the DADA2 method. Running dadasnake could not be easier: it is called by a single command from the command line. With a human-readable configuration file and a simple sample table, its steps are adjustable to a wide array of input data and requirements. It is designed to run on a computing cluster using a single conda environment in multiple jobs triggered by Snakemake. dadasnake reports on intermediary steps and statistics in intuitive figures and tables. Final data output formats include biom format, phyloseq objects, and flexible text files or R data sets for easy integration in microbial ecology analysis scripts.

Installing dadasnake

For dadasnake to work, you need conda.

1) Clone this repository to your disk:

git clone https://github.com/a-h-b/dadasnake.git

Change into the dadasnake directory:

cd dadasnake

At this point, you have all the scripts you need to run the workflow using snakemake, and you'd just need to get some data and databases (see point 8). If you want to use the comfortable dadasnake wrapper, follow the points 2-6.

2) Decide how you want to run dadasnake, if you let it submit jobs to the cluster: Only do one of the two:

• if you want to submit the process running snakemake to the cluster:

  cp auxiliary_files/dadasnake_allSubmit dadasnake
  chmod 755 dadasnake

• if you want to keep the process running snakemake on the frontend using tmux:

  cp auxiliary_files/dadasnake_tmux dadasnake
  chmod 755 dadasnake

If you don't submit jobs to the cluster, but want to run the whole workflow interactively, e.g. on a laptop, it doesn't matter which wrapper you use. Just copy one of them, as described above.

3) Adjust the file VARIABLE_CONFIG to your requirements (have a tab between the variable name and your setting):

• SNAKEMAKE_VIA_CONDA - set this to true, if you don't have snakemake in your path and want to install it via conda. Leave empty, if you don't need an additional snakemake.
• SNAKEMAKE_EXTRA_ARGUMENTS - if you want to pass additional arguments to snakemake, put them here (e.g. --latency-wait=320 for slower file systems). Leave empty usually.
• LOADING_MODULES - insert a bash command to load modules, if you need them to run conda. Leave empty, if you don't need to load a module.
• SUBMIT_COMMAND - insert the bash command you'll usually use to submit a job to your cluster to run on a single cpu for a few days. You only need this, if you want to have the snakemake top instance running in a submitted job. You alternatively have the option to run it on the frontend via tmux. Leave empty, if you want to use this frontend version and have tmux installed. You don't need to set this, if you are wanting to run the workflow interactively / on a laptop.
• BIND_JOBS_TO_MAIN - if you use the option to run the snakemake top instance in a submitted job and need to bind the other jobs to the same node, you can set this option to true. See FAQ below for more details. You don't need to set this, if you are wanting to run the workflow interactively / on a laptop.
• NODENAME_VAR - if you use the BIND_JOBS_TO_MAIN option, you need to let dadasnake know, how to access the node name (e.g.SLURMD_NODENAME on slurm). You don't need to set this, if you are wanting to run the workflow interactively / on a laptop.
• SCHEDULER - insert the name of the scheduler you want to use (currently slurm or uge). This determines the cluster config given to snakemake, e.g. the cluster config file for slurm is config/slurm.config.yaml . Also check that the settings in this file is correct. If you have a different system, contact us ( https://github.com/a-h-b/dadasnake/issues ). You don't need to set this, if you are wanting to run the workflow interactively / on a laptop.
• MAX_THREADS - set this to the maximum number of cores you want to be using in a run. If you don't set this, the default will be 50. Users can override this setting at runtime.
• NORMAL_MEM_EACH - set the size of the RAM of one core of your normal copute nodes (e.g. 8G). If you're not planning to use dadasnake to submit to a cluster, you don't need to set this.
• BIGMEM_MEM_EACH - set the size of the RAM of one core of your bigmem (or highmem) compute nodes. If you're not planning to use dadasnake to submit to a cluster or don't have separate bigmem nodes, you don't need to set this.
• BIGMEM_CORES - set this to the maximum number of bigmem cores you want to require for a task. Set to 0, if you don't have separate bigmem nodes. You don't need to set this, if you're not planning to use dadasnake to submit to a cluster.
• LOCK_SETTINGS - set this to true, if you don't want users to choose numbers and sizes of compute nodes at run time. If you're not planning to use dadasnake to submit to a cluster, you don't need to set this. Setting LOCK_SETTINGS makes the workflow slightly less flexible, as all large data sets will be run with the maximum number of bigmem nodes you set up here (see big_data settings below). On the other hand, it can be helpful, if you're setting up dadasnake for inexperienced users or have only one possible setting anyhow. If you're not locking, it's advised to set useful settings in the config/config.default.yaml file for normalMem, bigMem, and bigCores.

4) optional, but highly recommended: Install snakemake via conda: If you want to use snakemake via conda (and you've set SNAKEMAKE_VIA_CONDA to true), install the environment, as recommended by Snakemake:

conda install -c conda-forge mamba
mkdir -p conda
mamba create --prefix $PWD/conda/snakemake_env
conda activate $PWD/conda/snakemake_env
mamba install -c conda-forge -c bioconda snakemake=6.9.1 mamba tabulate=0.8
conda deactivate

Alternatively, if the above does not work, you can install a fixed snakemake version without mamba like so:

conda env create -f workflow/envs/snakemake_env.yml --prefix $PWD/conda/snakemake_env

Dadasnake will run with Snakemake version >= 5.9.1 and hasn't been tested with any previous versions.

5) Set permissions / PATH: Dadasnake is meant to be used by multiple users. Set the permissions accordingly. I'd suggest:

• to have read access for all files for the users plus
• execution rights for the dadasnake file and the .sh scripts in the subfolder submit_scripts
• read, write and execution rights for the conda subfolder
• Add the dadasnake directory to your path.
• It can also be useful to make the VARIABLE_CONFIG file not-writable, because you will always need it. The same goes for config.default.yaml once you've set the paths to the databases you want to use (see below).

6) Initialize conda environments: This run sets up the conda environments that will be usable by all users:

./dadasnake -i config/config.init.yaml 

This step will take several minutes. It will also create a folder with the name "dadasnake_initialized". You can safely remove it or keep it. I strongly suggest to remove one line from the activation script after the installation, namely the one reading: R CMD javareconf > /dev/null 2>&1 || true, because you don't need this line later and if two users run this at the same time it can cause trouble. You can do this by running:

sed -i "s/R CMD javareconf/#R CMD javareconf/" conda/*/etc/conda/activate.d/activate-r-base.sh

7) Optional test run: The test run does not need any databases. You should be able to start it by running

./dadasnake -l -n "TESTRUN" -r config/config.test.yaml

If all goes well, dadasnake will run in the current session, load the conda environment, and make and fill a directory called testoutput. A completed run contains a file "workflow.done". If you don't want to see dadasnake's guts at this point, you can also run this with the -c or -f settings to submit to your cluster or start a tmux session (see How to run dadasnake below).

8) Databases: The dadasnake does not supply databases. I'd suggest to use the SILVA database for 16S data and UNITE for ITS.

• dadasnake can use mothur to do the classification, as it's faster and likely more accurate than the legacy DADA2 option. You need to format the database like for mothur (see here).
• dadasnake can alternatively use the DADA2 implementation of the same classifier. You can find some databases maintained by Michael R. McLaren here. More information on the format is in the DADA2 tutorial.
• In addition to the bayesian classifier, dadasnake implements DECIPHER. You can find decipher databases on the decipher website or build them yourself.
• dadasnake can use fungal traits to assign traits to fungal genere. Download the latest table from here - dadasnake has been tested with v1.2.
• You can also use dadasnake to blast and summarize results using basta. Have a look at the NCBI's ftp.
• To annotate fungal taxonomy with guilds via funguild, if you have suitable databases.
• If you still have a tax4fun2 installation, you can also use it within dadasnake. The package and database were taken off github, so it's not part of defaut dadasnake anymore.
• You can now use picrust2 within dadasnake. You need to set the path to the databases of your choice in the config file. By default, dadasnake looks for databases in the directory above where it was called. It makes sense to change this for your system in the config.default.yaml file upon installation, if all users access databases in the same place.

9) Fasttree: dadasnake comes with fasttree for treeing, but if you have a decent number of sequences, it is likely to be relatively slow. If you have fasttreeMP, you can give the path to it in the config file.

How to cite dadasnake

Christina Weißbecker, Beatrix Schnabel, Anna Heintz-Buschart, Dadasnake, a Snakemake implementation of DADA2 to process amplicon sequencing data for microbial ecology, GigaScience, Volume 9, Issue 12, December 2020, giaa135. Please also cite DADA2: Callahan, B., McMurdie, P., Rosen, M. et al. DADA2: High-resolution sample inference from Illumina amplicon data. Nat Methods 13, 581–583 (2016), and any other tools you use within dadasnake, e.g. mothur, DECIPHER, ITSx, Fasttree, VSEARCH, FUNGuild, PICRUSt2, BASTA, tax4fun2.

How to run dadasnake

To run the dadasnake, you need a config file and a sample table, plus data:

• The config file (in yaml format) is read by Snakemake to determine the inputs, steps, arguments and outputs.
• The sample table (tab-separated text) always gives sample names and file names, with column headers named library and r1_file (and r2_file for paired-end data sets). The path to the sample table has to be mentioned in the config file. You can add columns labeled run and sample to indicate libraries that should be combined into one final column and different sequencing runs (see the section about the sample table below).
• All raw data (usually fastq files) need to be in one directory (which has to be given in the config file).
• It is possible (and the best way to do this) to have one config file per run, which defines all settings that differ from the default config file.

Using the dadasnake wrapper

As shown in the installation description above, dadasnake can be run in a single step, by calling dadasnake. Since most of the configuration is done via the config file, the options are very limited. You can either:

• -c run (submit to a cluster) dadasnake and make a report (-r), or
• -l run (in the current terminal) dadasnake and make a report (-r), or
• -f run (in a tmux session on the frontend) dadasnake only available in the tmux installation and make a report (-r), or
• just make a report (-r), or
• run a dryrun (-d), or
• unlock a working directory, if a run was killed (-u)
• initialize the conda environmnets only (-i) - you should only need this during the installation. It is strongly recommended to first run a dryrun on a new configuration, which will tell you within a few seconds and without submission to a cluster whether your chosen steps work together, the input files are where you want them, and your sample file is formatted correctly. In all cases you need the config file as the last argument.

  dadasnake -d -r config.yaml

  You can also set the number of cpus to maximally run at the same time with -t. The defaults (1 for local/frontend runs and 50 for clusters) are reasonable for many settings and if you don't know what this means, you probably don't have to worry. But you may want to increase the numbers for larger datasets or bigger infrastructure, or decrease the numbers to match your environment's constraints. You can add a name for your main job (-n NAME), e.g.:

  dadasnake -c -n RUNNAME -r config.yaml

  Note that spaces in RUNNAME are not allowed and dots will be replaced by underscores.

If you use the tmux version, you can see the tmux process running by typing tmux ls. You can also see the progress by checking the stdandard error file tail RUNNAME_XXXXXXXXXX.stderr.

Depending on your dataset and settings and your cluster's scheduler, the workflow will take a few minutes to days to finish.

Running snakemake manually

Once raw data, config file and sample file are present, the workflow can be started from the dadasnake directory by the snakemake command:

snakemake -s Snakefile --configfile /PATH/TO/YOUR/CONFIGFILE --use-conda

If you're using a computing cluster, add your cluster's submission command and the number of jobs you want to maximally run at the same time, e.g.:

snakemake -j 50 -s Snakefile --cluster "qsub -l h_rt={resources.runtime},h_vmem=8G -pe smp {threads} -cwd" --configfile /PATH/TO/YOUR/CONFIGFILE --use-conda 

This will submit most steps as their own job to your cluster's queue. The same can be achieved with a cluster configuration:

snakemake -j 50 -s Snakefile --cluster-config PATH/TO/SCHEDULER.config.yaml --cluster "{cluster.call} {cluster.runtime}{resources.runtime} {cluster.mem_per_cpu}{resources.mem} {cluster.threads}{threads} {cluster.partition}" --configfile /PATH/TO/YOUR/CONFIGFILE --use-conda

If you want to share the conda installation with colleagues, use the --conda-prefix argument of Snakemake

snakemake -j 50 -s Snakefile --cluster-config PATH/TO/SCHEDULER.config.yaml --cluster "{cluster.call} {cluster.runtime}{params.runtime} {cluster.mem_per_cpu}{resources.mem} {cluster.threads}{threads} {cluster.partition}" --use-conda --conda-prefix /PATH/TO/YOUR/COMMON/CONDA/DIRECTORY

Depending on your dataset and settings, and your cluster's queue, the workflow will take a few minutes to days to finish.

What does the dadasnake do?

• primer removal and removal of poly-G-tails - using cutadapt
• quality filtering and trimming - using DADA2
• optional downsampling of reads per sample - using seqtk
• error estimation & denoising - using DADA2, including Novaseq-enabled models
• paired-ends assembly - using DADA2
• "OTU" table generation (it contains ASVs, of course) - using DADA2
• chimera removal - using DADA2
• clustering of ASVs at a user-set similarity (these are called OTU now)
• taxonomic classification - using mothur and/or DECIPHER (& ITS detection - using ITSx & blastn + BASTA)
• functional annotation - using funguild, fungalTraits, picrust2 (or tax4fun2)
• length check - in R
• treeing - using clustal omega and fasttree
• hand-off in biom-format, as R object, as R phyloseq object, and as fasta and tab-separated tables
• keeping tabs on number of reads in each step, and read quality control - using fastqc & multiQC You can control the settings for each step in a config file.

The samples table

Every samples table needs sample names (under header library) and file names (just the names, the path should be in the config file under header r1_file and potentially r2_file). Since DADA2 estimates run-specific errors, it can be helpful to give run IDs (under header run). If you have many (>500 samples), it is also useful to split them into runs for the analysis, as some of the most memory-intensive steps are done by run.
If several fastq files should end up in the same column of the ASV/OTU table, you can indicate this by giving these libraries the same sample name (under header sample). Libraries from different runs are combined in the final ASV/OTU table (example 1). Libraries from the same run are combined after primer-processing (example 2). Example 1: Example 2:

The configuration

The config file must be in .yaml format. The order within the yaml file does not matter, but the hierarchy has to be kept. Explanations can be found in the config-file in config/config.default.yaml .

What if something goes wrong?

If you gave dadasnake your email address and your system supports mailing (to that address), you will receive an email upon start and if the workflow encountered a problem or after the successful run. If there was a problem, you have to check the output and logs.

• Use the -d option of dadasnake or the --dryrun option of Snakemake before the run to check that your input files are where you want them and that you have permissions to write to your target directory. This will also do some checks on the configuration and samples table, so it discovers the majority of errors on a suitable combination of dataset and configuration.
• You can not make two runs of dadasnake write to the same output directory. If you start the second run while the first is still running, you will get an error either indicating that the directory can't be locked, or that the metadata is incomplete. If you've finished the first run already, the dadasnake will tell you that there's nothing to be done. Change the output directory in the config file to be unique for each run.
• A common reason for errors are misformatted inputs, e.g. the databases for the classification or the read files.
• dadasnake should catch most errors related to empty outputs. For example: the filtering is too stringent and no sequences are left; the primers you expected to find are not present; the sequences were truncated too short to be merged. Please report issues where this didn't happen.
• The best way to pinpoint those errors is to first check the .stderr file made by dadasnake (or the Snakemake output, if you run the workflow outside dadasnake). This will tell you which rule encountered the error, and, if you use the cluster submission, the job ID. You may have to search for the error a bit, because dadasnake will try to finish as much as possible of your run before dying. Hint: you can find errors by colour or by searching for "Error in rule".
• If you use the cluster submission, log files for every rule are written into the output directory and you can check the one with the job ID for additional information, otherwise the same information is written to the Snakemake output.
• The logs directory in the output directory contains log files for all steps that can produce comments. They are named with the step and then the name of the rule, so you can check the log file of the step that sent the error. Depending on the tool that sent the error, this will be easy to understand or cryptic. Don't hesitate to raise an issue in this repository if you get stuck.

How to ...?

I don't have primers on my reads, what do I do? Set do_primers: false in the configuration file, but make sure that orientation of the reads is the same.

I did paired end sequencing, but my reads are too short to overlap You have two options: 1) use only one read (usually the first) by setting paired: false in the config file and providing only the read you want to use in the samples table. This will run a single-end workflow. The makers of DADA2 would probably recommend this option in most cases. 2) use both reads, set a truncation length for filtering to make sure the sequences have the same lengths and use DADA2's option to "merge" reads without overlap e.g.

filtering:
  trunc_length:
    fwd: 250
    rvs: 200
pair_merging:
  min_overlap: 0
  just_concatenate: true

I need to set further parameters for job submission You can change the cluster configs and add the parameter, for example directly as part of the call field.

I need to bind the jobs to the same node as the main job Yes, you can. If you use the submission-based wrapper, you can provide the flag for choosing a node as part of the SUBMIT_COMMAND variable in the VARIABLE_CONFIG file. Also, specify BIND_JOBS_TO_MAIN as true. You also need to set the variable that holds the node's name in your submission system as NODENAME_VAR. All jobs will then be submitted to the same node as the one that runs the main snakemake, if you include the flag for choosing a node as part of the call field in the cluster config. You can also specify that one, using -b. Example: VARIABLE_CONFIG file:

...
SUBMIT_COMMAND  slurm --nodelist=
BIND_JOBS_TO_MAIN   true
NODENAME_VAR    SLURMD_NODENAME
SCHEDULER   slurm_simple
...

slurm_simple.config:

__default__:
  call: "sbatch --nodelist="
  mem_per_cpu: "--mem-per-cpu "
  partition: ""
  runtime: "-t"
  threads: "-c"
  stdout: "-o dadasnake.{rule}.{wildcards}.stdout"

call:

./dadasnake -c -b favorite_node -n TESTRUN config/config.test.yaml

I have a very large dataset Great, if you have the computing power to match it, dadasnake will help you. It has successfully processed >27,000 samples in the same run. If you run out of memory in your run, set big_data to true in the config file and allow the use of multiple bigmem cores (we needed 360GB RAM for the 27,000 dataset). Disable highly memory intensive steps, such as treeing, chimera removal, plotting of rarefaction curves. If you didn't use the grouping by runs in your sample table, invent some runs of approx 100 samples each - these will be treated separately for some of the heavier DADA2 steps (error estimation).

How do I restart a failed run? Depends on why it failed...

• If you ran into a time limit or similar, you can just run dadasnake on the same config with the -u option and then again with the -c option. This will make Snakemake pick up where it left off.
• For most other situations, it's probably best to fix what caused the error in your config file and delete the output directory to start from scratch. If you're going to be loosing a lot of run time to that, and you're quite certain the problem is only in the last attempted step, you can try to restart. Ask us, if in doubt.

Can I restart from a certain step? If you're familiar with Snakemake, you can use it to force re-running the steps you need. It's not (yet) part of the dadasnake to do this more comfortably.