Research
Our group develops novel computational methods that combine big-data algorithms and machine learning to make sense of the ever-growing universe of biological sequences and structures. We build fast, open-source tools that search, cluster, assemble, and predict proteins and genomes at the scale of entire databases — turning massive genomic and proteomic datasets into biological insight, from microbial communities to the organization of the protein structure universe.
Sequence search & clustering
MMseqs2 (Many-against-Many sequence searching) is a software suite to search and cluster huge protein and nucleotide sequence sets, running orders of magnitude faster than BLAST while matching PSI-BLAST sensitivity in profile searches.
Linclust clusters huge sequence sets with a runtime that scales linearly with input size — not quadratically as in conventional algorithms — making it orders of magnitude faster than comparable methods.
Protein structure
Folddisco indexes and searches discontinuous structural motifs — such as catalytic sites and binding pockets whose residues are distant in sequence — enabling fast motif queries across millions of protein structures.
FoldMason builds accurate multiple structure alignments of proteins at scale, leveraging structural information to align thousands of remote homologs that elude sequence-based methods and to reconstruct their evolutionary relationships.
Foldseek-Multimer extends Foldseek to align and search protein complexes, enabling rapid and sensitive comparison of multi-chain structures across the entire predicted complex universe.
AFDB Clusters applies Foldseek clustering to the entire AlphaFold Protein Structure Database, grouping the predicted protein universe into structural clusters that reveal previously unseen protein families and their relationships.
Foldseek is a software suite for searching and clustering protein structures, orders of magnitude faster than state-of-the-art structural aligners — letting you query millions of structures in seconds.
ColabFold is an easy-to-use environment for fast protein structure prediction. It pairs AlphaFold2 and RoseTTAFold with fast MMseqs2-based MSA generation, making predictions orders of magnitude faster than the standard AlphaFold system.
Metagenomics
Metabuli is a metagenomic classification tool that jointly analyzes amino acid and DNA sequences, combining the sensitivity of amino-acid searches with the specificity of DNA-level resolution to taxonomically profile complex samples.
Conterminator is an efficient method for detecting incorrectly labeled sequences across kingdoms by an exhaustive all-against-all sequence comparison.
Plass assembles short-read sequencing data at the protein level, targeting complex metagenomes. It recovers many times more protein residues from soil metagenomes than nucleotide assemblers such as Megahit.