MetaDecoder

An algorithm for clustering metagenomic sequences.

Cite MetaDecoder:

Liu, CC., Dong, SS., Chen, JB. et al. MetaDecoder: a novel method for clustering metagenomic contigs. Microbiome 10, 46 (2022). https://doi.org/10.1186/s40168-022-01237-8

Please note that MetaDecoder will not be updated to optimize performance, only bug fixes. An updated version of MetaDecoder is already in development, which will bring more accurate clustering results and lower GPU resource requirements.

2023年，青基评委给我不予资助的最终意见，理由是所谓的没有创新性。我不确定MetaDecoder2能否最终上线，因为仍有许多与科研无关的事情在消磨我的精力。请对青年科研工作者一些宽容，加油吧，最后感谢基金委。 - 20230827

Benchmarks

All datasets mentioned in text and some NEWLY ADDED datasets are available in Google Drive.

Benchmarks for all datasets are available in benchmarks directory.

Dependencies

• python (version >= 3.8)
• numpy
• scipy
• scikit-learn
• threadpoolctl
• fraggenescan (version 1.31)
• hmmer (version 3.2.1)

Installation

Download and install MetaDecoder (Do not clone this repository)

# You may need to install pip3 before. #
curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
python3 get-pip.py

# You may need to install or upgrade setuptools and wheel using pip3 before. #
pip3 install --upgrade setuptools wheel

# Download and install MetaDecoder version 1.1.0 #
pip3 install -U https://github.com/liu-congcong/MetaDecoder/releases/download/v1.1.0/metadecoder-1.1.0-py3-none-any.whl

Make sure you have a good internet connection, as MetaDecoder will install the required Python dependencies, you can also install the dependencies manually:

• numpy
• scipy
• scikit-learn
• threadpoolctl

pip3 install numpy scipy scikit-learn threadpoolctl

MetaDecoder uses FragGeneScan and Hmmer for predicting protein coding genes and mapping single-copy marker genes to contigs, respectively.

MetaDecoder has included the compiled FragGeneScan (version 1.31) and Hmmer (version 3.2.1).

The GPU version of MetaDecoder

MetaDecoder can be accelerated using the GPU on the basis of CUDA and CuPy.

To use the GPU version of Metadecoder, you need to have a compatible driver installed for your GPU (CUDA), and then install CuPy using pip3:

# You may need to install and upgrade setuptools and wheel using pip3 before. #
pip3 install --upgrade setuptools wheel
# Please note that XXX is the CUDA version. e.g. cupy-cuda101 means CuPy with CUDA 10.1. #
pip3 install cupy-cudaXXX

Please be careful not to install multiple CuPy packages at the same time.

MetaDecoder will automatically enable GPU if it is available.

And moreover, CuPy can use additional CUDA library (cuTENSOR) to accelerate tensor operations: UNTESTED

# Please note that XXX is the CUDA version. #
python3 -m cupyx.tools.install_library --cuda XXX --library cutensor
# Setting the environment variable to activate some CUDA features in CuPy. #
echo 'export CUPY_ACCELERATORS="cutensor"' >> ~/.bashrc

Usage

Preparations

Before running MetaDecoder, you may need to prepare some files by yourself.

• A FASTA formatted assembly file: ASSEMBLY.FASTA

• Some sorted BAM formatted read files with the SAME HEADER: SAMPLE1.BAM, SAMPLE2.BAM ...

Run MetaDecoder

Obtain the coverages of contigs

Input: SAMPLE1.BAM, SAMPLE2.BAM, ...

Output: METADECODER.COVERAGE

metadecoder coverage -b SAMPLE1.BAM SAMPLE2.BAM ... -o METADECODER.COVERAGE

Since v1.0.17, MetaDecoder supports calculating coverage for each sample, which can reduce the storage space for multiple sequencing samples.

Since v1.1.0, MetaDecoder can work with sorted BAM files and does not support SAM files.

for file in *.bam
do
metadecoder coverage -s ${file}.bam -o ${file}.METADECODER.COVERAGE
done

Map single-copy marker genes to the assembly

Input: ASSEMBLY.FASTA

Output: METADECODER.SEED

metadecoder seed --threads 50 -f ASSEMBLY.FASTA -o METADECODER.SEED

Run MetaDecoder algorithm to cluster contigs

Input: ASSEMBLY.FASTA, METADECODER.COVERAGE, METADECODER.SEED

Output: METADECODER.1.FASTA, METADECODER.2.FASTA, ...

metadecoder cluster -f ASSEMBLY.FASTA -c METADECODER.COVERAGE -s METADECODER.SEED -o METADECODER

Since v1.0.17, MetaDecoder can load multiple coverage files for clustering.

metadecoder cluster -f ASSEMBLY.FASTA -c *.METADECODER.COVERAGE -s METADECODER.SEED -o METADECODER

A simple example to use MetaDecoder is available in MetaDecoder/example/

Change logs

• 1.0.3 (20211008): Initial version.

• 1.0.4 (20211029): Optimize the calculation of distance of pairwise kmer frequency.

• 1.0.5 (20211105): Optimize the counting process of kmers.

• 1.0.6 (20211207): Added an option (--no_clusters) to output only sequence IDs and the corresponding cluster IDs instead of sequences.

• 1.0.7 (20220206): Minor bugs fixed.

• 1.0.8 (20220315): Minor bugs fixed.

• 1.0.9 (20220508): Fix a bug that causes abnormal coverage when the "--mapq" parameter is set to 0.

• 1.0.10 (20220514): Minor bugs fixed.

• 1.0.11 (20220518): Minor bugs fixed.

• 1.0.12 (20220708): Minor bugs fixed.

• 1.0.13 (20220712): Minor bugs fixed.

• 1.0.14 (20220914): Minor bugs fixed.

• 1.0.15 (20221103): Minor bugs fixed.

• 1.0.16 (20221117): Minor bugs fixed.

• 1.0.17 (20230418): Multiple coverage files for clustering.

• 1.0.18 (20230816): Support gz formatted assemblies.

• 1.0.19 (20240125): Clusters will not participate in the calculation of the average of kmer distance if it contains more than 50,000 sequences.

• 1.1.0 (20241207): MetaDecoder has been updated to support sorted BAM files only, discontinuing support for SAM files [Thanks for @jolespin].

References

• Mina Rho, Haixu Tang, and Yuzhen Ye. FragGeneScan: Predicting Genes in Short and Error-prone Reads. Nucl. Acids Res., 2010 doi: 10.1093/nar/gkq747.

• nhmmer: DNA Homology Search With Profile HMMs. T. J. Wheeler, S. R. Eddy. Bioinformatics, 29:2487-2489, 2013.