Plass (Protein-Level ASSembler) and PenguiN (Protein guided nucleotide assembler) are software to assemble protein sequences or DNA/RNA contigs from short read sequencing data meant to work best for complex metagenomic or metatranscriptomic datasets. Plass and Penguin are GPL-licensed open source software implemented in C++ and available for Linux and macOS and are designed to run on multiple cores.

Plass: Steinegger M, Mirdita M and Soeding J. Protein-level assembly increases protein sequence recovery from metagenomic samples manyfold. Nature Methods, doi: doi.org/10.1038/s41592-019-0437-4 (2019).

PenguiN: Jochheim A, Jochheim FA, Kolodyazhnaya A, Morice E, Steinegger M, Soeding J. Strain-resolved de-novo metagenomic assembly of viral genomes and microbial 16S rRNAs. Microbiome 12, 187, (2024)

SRC was created by assembling 640 soil metagenome samples. MERC was assembled from the the metatranscriptomics datasets created by the TARA ocean expedition. Both catalogues were redundancy reduced to 90% sequence identity at 90% coverage. Each catalog is a single FASTA file containing the sequences, the header identifiers contain the Sequence Read Archive (SRA) identifiers. The catalogues can be downloaded here. We provide a HH-suite3 database called "BFD" containing sequences from the Metaclust, SRC, MERC and Uniport at here.

PenguiN a software to assemble short read sequencing data on a nucleotide level. In a first step it assembles coding sequences using the information from the translated protein sequences. In a second step it links them across non-coding regions. The main purpose of PenguiN is the assembly of complex metagenomic and metatranscriptomic datasets. It was especially tested for the assembly of viral genomes as well as 16S rRNA gene sequences. It assembles 3-40 times more complete viral genomes and six times as many 16S rRNA sequences than state of the art assemblers like Megahit and the SPAdes variants.

Our software can be install via conda or as statically compiled binaries. It requires a 64-bit Linux or macOS system.

 # install from bioconda
 conda install -c conda-forge -c bioconda plass 
 # install docker
 docker pull ghcr.io/soedinglab/plass:latest
 # static build with AVX2 (fastest)
 wget https://mmseqs.com/plass/plass-linux-avx2.tar.gz; tar xvfz plass-linux-avx2.tar.gz; export PATH=$(pwd)/plass/bin/:$PATH
 # static build with SSE4.1
 wget https://mmseqs.com/plass/plass-linux-sse41.tar.gz; tar xvfz plass-linux-sse41.tar.gz; export PATH=$(pwd)/plass/bin/:$PATH
 # universal build with macOS (Intel or Apple Silicon)
 wget https://mmseqs.com/plass/plass-osx-universal.tar.gz; tar xvfz plass-osx-universal.tar.gz; export PATH=$(pwd)/plass/bin/:$PATH

Other precompiled binaries for SSE2, ARM and PowerPC can be found at mmseqs.com/plass.

Plass and PenguiN can assemble both paired-end reads (FASTQ) and single reads (FASTA or FASTQ):

  # assemble paired-end reads 
  plass assemble examples/reads_1.fastq.gz examples/reads_2.fastq.gz assembly.fas tmp

  # assemble single-end reads 
  plass assemble examples/reads_1.fastq.gz assembly.fas tmp

  # assemble single-end reads using stdin
  cat examples/reads_1.fastq.gz | plass assemble stdin assembly.fas tmp

Important parameters:

 --min-seq-id         Adjusts the overlap sequence identity threshold
 --min-length         minimum codon length for ORF prediction (default: 40)
 -e                   E-value threshold for overlaps 
 --num-iterations     Number of iterations of assembly
 --filter-proteins    Switches the neural network protein filter off/on

Plass workflows:

  plass assemble      Assembles proteins (i:Nucleotides -> o:Proteins)

PenguiN workflows:

  penguin guided_nuclassemble  Assembles nucleotides using protein and nucleotide information (i:Nucleotides -> o:Nucleotides)
  penguin nuclassemble         Assembles nucleotides using only nucleotdie information (i:Nucleotides -> o:Nucleotides)

Both tools can be distributed over several homogeneous computers. However the tmp folder has to be shared between all nodes (e.g. NFS). The following command assembles on several nodes:

RUNNER="mpirun -np 42" plass assemble examples/reads_1.fastq.gz examples/reads_2.fastq.gz assembly.fas tmp

Compiling from source has the advantage that it will be optimized to the specific system, which should improve its performance. To compile git, g++ (4.9 or higher) and cmake (3.0 or higher) are required. Afterwards, the PLASS and PenguiN binaries will be located in the build/bin directory.

  git clone https://github.com/soedinglab/plass.git
  cd plass
  git submodule update --init
  mkdir build && cd build
  cmake -DCMAKE_BUILD_TYPE=RELEASE -DCMAKE_INSTALL_PREFIX=. ..
  make -j 4 && make install
  export PATH="$(pwd)/bin/:$PATH"

❗ If you want to compile PLASS or PenguiN on macOS, please install and use gcc from Homebrew. The default macOS clang compiler does not support OpenMP and PLASS will not be able to run multithreaded. Use the following cmake call:

  CXX="$(brew --prefix)/bin/g++-13" cmake -DCMAKE_BUILD_TYPE=RELEASE -DCMAKE_INSTALL_PREFIX=. ..

When compiling from source, our sofwtare requires the zlib and bzip installed.

We also provide a Docker image of Plass. You can mount the current directory containing the reads to be assembled and run plass with the following command:

  docker run -ti --rm -v "$(pwd):/app" -w /app ghcr.io/soedinglab/plass:latest assemble reads_1.fastq reads_2.fastq assembly.fas tmp

Plass needs roughly 1 byte of memory per residue to work efficiently. Plass will scale its memory consumption based on the available main memory of the machine. Plass needs a CPU with at least the SSE4.1 instruction set to run.

• The assembly of Plass includes all ORFs having a start and end codon that includes even very short ORFs < 60 amino acids. Many of these short ORFs are spurious since our neural network cannot distingue them well. We would recommend to use other method to verify the coding potential of these. Assemblies above 100 amino acids are mostly genuine protein sequences.
• Plass in default searches for ORFs of 40 amino acids or longer. This limits the read length to > 120. To assemble this protein, you need to lower the --min-length threshold. Be aware using short reads (< 100 length) might result in lower sensitivity.