MMseqs2: sensitive protein sequence searching for the analysis of massive data sets

By Martin Steinegger, Johannes Söding. 2017

1,091 citations recorded by the source corpus.

Sequencing costs have dropped much faster than Moore's law in the past decade, and sensitive sequence searching has become the main bottleneck in the analysis of large (meta)genomic datasets. While previous methods sacrificed sensitivity for speed gains, the parallelized, open-source software MMseqs2 overcomes this trade-off: In three-iteration profile searches it reaches 50% higher sensitivity than BLAST at 83-fold speed and the same sensitivity as PSI-BLAST at 270 times its speed. MMseqs2 therefore offers great potential to increase the fraction of annotatable (meta)genomic sequences.

Subfields: Ecology, Molecular Biology, Computational Biology & Bioinformatics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Ecology

In Molecular Biology

In Computational Biology & Bioinformatics

Citations in the corpus

In silico protein interaction screening uncovers DONSON’s role in replication initiation (2023, methodology): “Multiple sequence alignments (MSAs) and template inputs to the AlphaFold network were generated within the Colabfold pipeline which routes protein sequences to another server running the MSA software MMseqs2 (63).”
Evolutionary-scale prediction of atomic level protein structure with a language model (2022, neutral): “[48] Martin Steinegger and Johannes SÈ oding.”
Analysis of several key factors influencing deep learning-based inter-residue contact prediction (2019, methodology): “After CASP13, we updated the alignment generation pipeline MULTICOM-CLUSTER to include an extra search on a large metagenomic protein sequence database Metaclust50 that was built from 1.6 billion metagenomic sequence fragments with 50% sequence identity (Steinegger and Soding, 2018).”
Generating functional protein variants with variational autoencoders (2020, methodology): “In order 120 to avoid highly similar sequences occurring in the training and validation sets, we first clustered 121 all the sequences using mmseqs2 [35], and then added clusters chosen at random to the validation 122 set until the total number of sequences in the validation clusters reached 20% of ”
NetSurfP-2.0: improved prediction of protein structural features by integrated deep learning (2018, methodology): “The first exploits the HH-suite software (25), that runs quickly on individual sequences, while the second uses the MMseqs2 software (26), that is optimized for searches on large data sets.”
Clustering huge protein sequence sets in linear time (2017, methodology): “Linclust has been integrated into our free GPLv3-licensed MMseqs2 software suite [10].”
ProteinNet: a standardized data set for machine learning of protein structure (2019, methodology): “ for this step as JackHMMER is unable to perform MSA-to-MSA searches, but the MSAs used are the original, JackHMMER-derived ones. Based on the HHblits alignment scores, we cluster MSAs using MMseqs2 (Steinegger and Söding, 2017, 2) with the sought sequence identity threshold, an e-value threshold of”
Advances and Applications in the Quest for Orthologs (2019, methodology): “Perhaps the most straightforward solution is to replace BLAST, which is used by many graph-based methods, by a much faster homology search tools such as MMseqs2 (Steinegger and Söding 2017) or DIAMOND (Buchfink et al.”
Protein-to-genome alignment with miniprot (2022, methodology): “We ran GeMoMa with MMseqs2 (Steinegger and Söding, 2017) as the underlying engine.”