CAMP Normalization

Overview

This module is designed to function as both a standalone MAG Normalization pipeline as well as a component of the larger CAMP metagenomics analysis pipeline. As such, it is both self-contained (ex. instructions included for the setup of a versioned environment, etc.), and seamlessly compatible with other CAMP modules (ex. ingests and spawns standardized input/output config files, etc.).

The CAMP normalization pipeline provides normalization methods for count data in feature tables. The module outputs normalized taxonomic profiles applying the most widely used normalization approaches, including compositionality-aware transformations and scaling techniques.

Installation

1. Clone repo from Github.

git clone https://github.com/MetaSUB-CAMP/camp_normalization

2. Set up the conda environment (contains Snakemake, Click, and other essentials) using configs/conda/normalization.yaml.

# Create and activate conda environment 
cd camp_normalization
conda env create -f configs/conda/normalization.yaml
conda activate normalization

3. Update the relevant parameters (if applicable- for example, location of external non-conda tools) in test_data/parameters.yaml.

4. Make sure the installed pipeline works correctly.

# Run tests on the included sample dataset
python /path/to/camp_normalization/workflow/normalization.py test

Using the Module

Input: /path/to/samples.csv provided by the user.

Output: 1) An output config file summarizing 2) the module's outputs.

• /path/to/work/dir/normalization/final_reports/samples.csv for ingestion by the next module (ex. quality-checking)

Module Structure

└── workflow
    ├── Snakefile
    ├── normalization.py
    ├── utils.py
    ├── __init__.py
    └── ext/
        └── scripts/

• workflow/normalization.py: Click-based CLI that wraps the snakemake and other commands for clean management of parameters, resources, and environment variables.
• workflow/Snakefile: The snakemake pipeline.
• workflow/utils.py: Sample ingestion and work directory setup functions, and other utility functions used in the pipeline and the CLI.
• ext/: External programs, scripts, and small auxiliary files that are not conda-compatible but used in the workflow.

Running the Workflow

1. Make your own samples.csv based on the template in configs/samples.csv. Sample test data can be found in test_data/.

   • For example, ingest_samples in workflow/utils.py expects a feature table in csv format along with a metadata table. The column that contains a batch variable must be renamed to batch for the batch correction methods to run successfully.
   • samples.csv requires either absolute paths or paths relative to the directory that the module is being run in.

2. Update the relevant parameters in configs/parameters.yaml: choose the normalization methods either from the list of methods or specify the groups of methods that need to be applied.

3. Update the computational resources available to the pipeline in configs/resources.yaml.

Command Line Deployment

To run CAMP on the command line, use the following, where /path/to/work/dir is replaced with the absolute path of your chosen working directory, and /path/to/samples.csv is replaced with your copy of samples.csv. - The default number of cores available to Snakemake is 1 which is enough for test data, but should probably be adjusted to 10+ for a real dataset. - Relative or absolute paths to the Snakefile and/or the working directory (if you're running elsewhere) are accepted! - The parameters and resource config YAMLs can also be customized.

python /path/to/camp_normalization/workflow/normalization.py \
    (-c number_of_cores_allocated) \
    (-p /path/to/parameters.yaml) \
    (-r /path/to/resources.yaml) \
    -d /path/to/work/dir \
    -s /path/to/samples.csv

Slurm Cluster Deployment

To run CAMP on a job submission cluster (for now, only Slurm is supported), use the following. - --slurm is an optional flag that submits all rules in the Snakemake pipeline as sbatch jobs. - In Slurm mode, the -c flag refers to the maximum number of sbatch jobs submitted in parallel, not the pool of cores available to run the jobs. Each job will request the number of cores specified by threads in configs/resources/slurm.yaml.

sbatch -J jobname -o jobname.log << "EOF"
#!/bin/bash
python /path/to/camp_normalization/workflow/normalization.py --slurm \
    (-c max_number_of_parallel_jobs_submitted) \
    (-p /path/to/parameters.yaml) \
    (-r /path/to/resources.yaml) \
    -d /path/to/work/dir \
    -s /path/to/samples.csv
EOF

Finishing Up

1. To quality-check the processed FastQs, download and compare the collated MultiQC reports, which can be found at /path/to/work/dir/short_read_qc/final_reports/*_multiqc_report/html. Multiple rounds of preprocessing may be needed to fully get rid of low-quality bases, adapters, and duplicated sequences.

   • For example, the dataset I worked with required an additional round of fastp to trim 10 low-quality bases from the 5' and 4 low-quality bases from the 3' end respectively.
   • I recommend creating a new directory, which I've called /path/to/work/dir/short_read_qc/5_retrimming and placing reprocessed reads inside them.
   • Afterwards, I reran FastQC and MultiQC and collated summary statistics (ex. numbers of reads, etc.) from the reprocessed datasets manually. I also updated the location of the reprocessed reads in /path/to/work/dir/short_read_qc/final_reports/samples.csv to /path/to/work/dir/short_read_qc/5_retrimming.

2. To plot grouped bar graph(s) of the number of reads and bases remaining after each quality control step in each sample, set up the dataviz environment and follow the instructions in the Jupyter notebook:

conda env create -f configs/conda/dataviz.yaml
conda activate dataviz
jupyter notebook &

3. After checking over final_reports/ and making sure you have everything you need, you can delete all intermediate files to save space.

python3 /path/to/camp_normalization/workflow/normalization.py cleanup \
    -d /path/to/work/dir \
    -s /path/to/samples.csv

4. If for some reason the module keeps failing, CAMP can print a script containing all of the remaining commands that can be run manually.

python3 /path/to/camp_normalization/workflow/normalization.py --dry_run \
    -d /path/to/work/dir \
    -s /path/to/samples.csv

Credits

• This package was created with Cookiecutter as a simplified version of the project template.

• Free software: MIT License

• Documentation: https://camp-documentation.readthedocs.io/en/latest/normalization.html