SemiBin

SemiBin is a metagenomic binning tool (MAG builder). It is based on deep contrastive learning to incorporate background information (from reference genomes).

It achieves better results than other tools across a range of microbial habitats (both host-associated and environmental habitats).

SemiBin2 introduced self-supervised contrastive learning (removing the need for the more expensive semi-supervised step) and added support for binning long-read assemblies, making it both faster and more accurate than the original version.

Latest release

The current version is SemiBin 2.6.0 (released October 14, 2026). It raises the minimum requirements to Python 3.11, pandas 3, and PyTorch 2.4 (pixi and conda handle this for you; if you are stuck on Python 3.10 or pandas 2, stay on SemiBin 2.5.0).

It adds shell tab completion of subcommands and options. To enable it in bash or zsh, add this to your ~/.bashrc (or ~/.zshrc):

eval "$(register-python-argcomplete SemiBin2)"

Static completion scripts for bash, zsh, fish, tcsh, and PowerShell can also be generated, see the installation docs.

Most of the bug fixes make SemiBin stop early with a clear message where it previously failed late with an obscure error: multi_easy_bin now checks up front that every sample has at least one contig long enough for the must-link threshold, bin_long validates its arguments like bin does, and failures of prodigal, bedtools, or hmmsearch now report the exit status, the input, and the tool's own error output. Loading models saved by SemiBin2 versions before 2.2.0 works again with PyTorch 2.6 or newer, and crashes on small or low-quality assemblies where no marker genes pass the filters are fixed. See the ChangeLog for the full list of changes.

SemiBin is available on bioconda and PyPI. We recommend installing it with pixi, which can even run it without installing:

pixi exec -c bioconda -c conda-forge -s semibin SemiBin2

Alternatively, with conda:

conda install -c conda-forge -c bioconda semibin

SemiBin links

• SemiBin documentation
• SemiBin Github repository
• SemiBin users mailing-list
• SemiBin1 manuscript: Shaojun Pan et al. A deep siamese neural network improves metagenome-assembled genomes in microbiome datasets across different environments in Nature Communications 2022
• SemiBin2 manuscript: Shaojun Pan et al. SemiBin2: self-supervised contrastive learning leads to better MAGs for short- and long-read sequencing in Bioinformatics 2023