MSA search on the platform

MSA search builds a multiple sequence alignment (MSA) for each protein chain you give it. It runs MMseqs2 against UniRef30 (release 2302) and ColabFoldDB (release 202108), the databases the public ColabFold server uses, on Ivy's own infrastructure. Your sequences never leave Ivy. Each distinct chain comes back as an .a3m file with its alignment depth.

This is the same search Chai-1 and Boltz-2 run before they fold. Running msa_search needs a signed-in account; it is not part of the guest tool set.

What MSA search does here

• Aligns each chain on its own. 1 to 32 distinct protein chains per job. A chain you list twice is searched once.
• Returns unpaired alignments. Each chain gets its own alignment. No cross-chain pairing is built for complexes.
• Reports depth per chain: the number of sequences in each alignment, query included.
• Caches by organization. An alignment is kept for 30 days after its last use, inside your organization only, and Chai-1 and Boltz-2 read the same cache. A chain searched by any of the three is not searched again while it is cached.
• Runs as an async job. A chain that is already cached comes back in seconds. Chains that need a fresh search take longer, plus up to about 7 minutes when the search service has to start cold.

When to use it

• You want to inspect alignment depth or conservation for a chain, for example to judge whether a fold will have enough evolutionary signal behind it.
• You want to export the alignments (.a3m) for analysis elsewhere.
• You are about to fold a batch of complexes that share chains, and want to warm the cache first so each fold starts without waiting for its search.

When to use something else

Limits to know before you run

• Protein only. Sequences must be one-letter amino-acid codes; digits, gap characters and SMILES are refused. A nucleotide sequence written in A/C/G/T letters is not caught, because those are amino-acid letters too, so check that you are passing protein.
• At most 32 distinct chains per job, each up to 5,000 residues.
• No cross-chain pairing. Alignments are per chain, so a paired MSA for a heterodimer is not available here.
• Not available to guests. The cache is partitioned by organization (or by user for an account with no organization), so a guest session has nowhere to keep alignments and is refused.
• Cost. 1 credit for the job, plus metered search time for chains that were not cached, typically 2–7 credits per search. Cached chains are free, and so is the cold start of the search service.

Citation

Mirdita M, Schütze K, Moriwaki Y, Heo L, Ovchinnikov S, Steinegger M. ColabFold: making protein folding accessible to all. Nature Methods 19, 679–682 (2022). doi: 10.1038/s41592-022-01488-1

Steinegger M, Söding J. MMseqs2 enables sensitive protein sequence searching for the analysis of massive data sets. Nature Biotechnology 35, 1026–1028 (2017). doi: 10.1038/nbt.3988

Source code: soedinglab/MMseqs2.

Read next