Why Use This Engine?
RevMSA builds the multiple sequence alignment (MSA) a structure prediction model needs before it can predict a protein. You enter the protein chains of one target or complex. RevMSA finds each chain’s related sequences in large public protein databases and lines them up against your chain, residue by residue. It returns the result as input files for Boltz, Chai-1 and ColabFold, so you can run those models without an alignment step of their own. Use it to prepare inputs for structure predictions you run yourself, to keep your sequences inside Revilico rather than sending them to a public alignment server, and to give several models the same alignment so their predictions are comparable. It is also a quick check before a prediction: how many relatives each chain has, which stretches have few, and whether the chains of a complex have relatives in the same organisms.Background
A multiple sequence alignment stacks related proteins under your sequence so that matching positions sit in the same column. Each row is one relative. Where a relative lacks a residue, the alignment has a gap. Where it has extra residues that your sequence does not, those are insertions. Structure prediction models read evolutionary information from the alignment. Positions that never change across hundreds of relatives are usually buried or functional, and positions that change together are often in contact. The number of relatives covering a position is its coverage. Where coverage is low, the model has little information to work with, and its prediction there is least reliable. For a complex of two or more different chains, RevMSA also builds a paired alignment. Each of its rows holds relatives of the different chains from the same organism, side by side. This is how a model sees which positions of one chain change together with positions of the other, which points to where the chains touch. A single chain, or a homodimer entered as two identical chains, has no paired alignment.Running the Engine
Inputs
Enter the chains of one target or complex. Paste them into Protein sequences, or upload a file.
A run takes 1 to 20 chains of 10 to 4,000 residues each, from files up to 1 MB. Click Load example to try it with human hemoglobin alpha and beta.
There are no search settings to choose.
Preparing your sequences
Enter the chains in the order you will use them in your structure model. The Boltz and ColabFold files are numbered in that order,seq_0 for the first chain, seq_1 for the second, and so on. Enter a chain once for each copy in the complex. A repeated chain is searched once, but each copy keeps its own numbered file.
Sequences must be protein, in one-letter amino acid codes, with X for an unknown residue. Spaces, line breaks and position numbers are removed, so a sequence copied from UniProt or GenBank works as pasted. A DNA or RNA sequence is refused. Enter the translated protein instead.
Pasted chains are checked as you type, and the list shows any problem next to the chain. You can cancel a queued or running alignment from its card in Command Center.