Allow to specify the number of the first residue of a chain

[lucajovine]

Something that I find really inconvenient in AF3 is the lack of an option to specify what number should be assigned to the first residue of a given chain (if the user does not want it to be the default "1"). I have tried convincing the AF3 team to introduce this but they did not want to do so, as in principle this can of course also be done using postprocessing scripts. Nonetheless, I do think that implementing this directly would make people's life easier, so hopefully there is room for such an enhancement in OF3?

Once again using the example examples/example_inference_inputs/query_multimer.json as a template, what I am suggesting is something like this:

{
    "queries": {
        "7cnx": {
            "chains": [
                {
                    "molecule_type": "protein",
                    "chain_ids": [
                        "A",
                        "C"
                    ],
                    "sequence": "MLNSFKLSLQYILPKLWLTRLAGWGASKRAGWLTKLVIDLFVKYYKVDMKEAQKPDTASYRTFNEFFVRPLRDEVRPIDTDPNVLVMPADGVISQLGKIEEDKILQAKGHNYSLEALLAGNYLMADLFRNGTFVTTYLSPRDYHRVHMPCNGILREMIYVPGDLFSVNHLTAQNVPNLFARNERVICLFDTEFGPMAQILVGATIVGSIETVWAGTITPPREGIIKRWTWPAGENDGSVALLKGQEMGRFKLG",
					"starting_residue_number:": "102"
                },
                {
                    "molecule_type": "protein",
                    "chain_ids": [
                        "B",
                        "D"
                    ],
                    "sequence": "XTVINLFAPGKVNLVEQLESLSVTKIGQPLAVSTGHHHHHHG",
					"starting_residue_number:": "22"
                }
            ]
        }
    }
}

which would generate PDBs where chains A and C would contain residues M102-G354, and chains B and D would contain residues X22-G63.

Thank you!!!

[smlewis]

I'll just commiserate and note I've never seen a piece of biophysics software (including the ones I worked on) treat numbering in the way anyone wanted it treated. Did we somehow solve protein folding before protein numbering?

[lucajovine]

Eh eh, indeed! So let's show AF3 that OF3 can do the right thing :-)

[jnwei]

@gnikolenyi do you have thoughts about how difficult this might be to implement? do we have downstream assumptions about the numbering of each chain starting from a specific value?

[lucajovine]

As a follow-up, I understand that some specific cases entail rather complicated numbering schemes (to account for insertions, for example, etc.). But I think the ability to just just shift the sequence numbering by n would take care of 99% of the cases one normally deals with.
To give an example, my group works on secreted proteins that generally have N-terminally signal peptides, which are removed as they enter the ER and thus not present in the mature proteins. The latter may thus start with something like Q25 or N19... you get the picture. Being able to feed only the mature sequences to OF3 (which of course one can already do!) AND tell it that the starting residues are number 25 or 19, rather than 1, would simplify everything else that happens downstream.
Another possible scenario is modeling the effect of modeling the addition of an N-terminal tag to a certain protein, while keeping the numbering of the resulting fusion consistent with that of the native protein.
In both these cases, simply being able to define a numerical offset for the first residue would help tremendously. On the other hand, I do not think it is really needed to have an even more general treatment that includes much more complicated numbering schemes, because those are relatively rare and could then be tackled ad hoc at the postprocessing stage.

[smlewis]

@decarboxy pointed out to me that CIF supports author defined IDs https://mmcif.wwpdb.org/docs/tutorials/content/atomic-description.html

[lucajovine]

Indeed. However, without getting into the mine field of how many structural biologists actually use CIF on a daily basis for anything beyond simply visualizing coordinates, I would argue that numbering residue 25 as 25 is not an author-defined ID. Instead, it is (mis-)numbering it as 1 that is an AlphaFold-defined ID! At some point, biology should be taken into account...

[chodec]

Hey @lucajovine, could you assign me this issue? I'll try my best to implement this, thanks

[lucajovine]

@chodec I don't think I have the permission to do so, sorry :-(

[jnwei]

Thanks @chodec for offering to work on this!

A couple of pointers that might help with implementing this feature:

• Because the model makes assumptions about the numbering of indexing of the residues in a chain, it is probably easiest to apply renumbering as a posprocessing step. That is, allow the OF3OutputWriter to handle the final numbering of the residues after predictions are complete. It would be necessary to pass in the new numbering scheme as part of the feature dict so that it can be received in OF3OutputWriter

• Another option that would support complex numbering schemes (insertions, deletions as @lucajovine mentioned) is to pass in a list of residue ids with the expected numbering for all residues, rather than just the initial offset, so perhaps something like this:

    "queries": {
        "7cnx": {
            "chains": [
                {
                    "molecule_type": "protein",
                    "chain_ids": [
                        "A",
                        "C"
                    ],
                    "sequence": "MLNSFKLSLQYILPKLWLTRLAGWGASKRAGWLTKLVIDLFVKYYKVDMKEAQKPDTASYRTFNEFFVRPLRDEVRPIDTDPNVLVMPADGVISQLGKIEEDKILQAKGHNYSLEALLAGNYLMADLFRNGTFVTTYLSPRDYHRVHMPCNGILREMIYVPGDLFSVNHLTAQNVPNLFARNERVICLFDTEFGPMAQILVGATIVGSIETVWAGTITPPREGIIKRWTWPAGENDGSVALLKGQEMGRFKLG",
		  "residue_ids:": ["102", "103", "104", ...]
                },
                {
                    "molecule_type": "protein",
                    "chain_ids": [
                        "B",
                        "D"
                    ],
                    "sequence": "XTVINLFAPGKVNLVEQLESLSVTKIGQPLAVSTGHHHHHHG",
                     "residue_ids:": ["22", "23", "110", "111", "112", "24", ...]   # arbitrary example of an insertion 
                }
            ]
        }
    }
}

You can have a convenience function that could convert a starting_residue_nubmer value into a residue_ids list.

Let us know if we can provide any more pointers / answer any other questions!

[lucajovine]

Thanks @jnwei (and @chodec)!

Re: applying renumbering as a posprocessing step, I agree — as long as that won't cause inconsistencies with indexes used for PAE plots etc. (will it?)

Re: complex numbering schemes, what you suggested would in principle be ideal as it would take care of all possible scenarios, even rare ones. But clearly a convenience function that (internally) converts a starting_residue_number value into a residue_numbers list would be essential, as I would not expect regular users to be happy about numbering each residue of a 3,000 aa input!
And finally, I would also make it so, that if starting_residue_number or a residue_numbers list are not specified, the system defaults to numbering from 1 (as it does now). But this is obvious!