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 2.4.1 (released August 2026). This is a small bugfix release: single-sample runs (single_easy_bin and generate_sequence_features_single) now check upfront whether any input contig is at least as long as the must-link threshold and, if not, abort immediately with an informative message instead of failing later with an obscure error. See the ChangeLog for the full history.

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