colabfold_batch still runs a MSA query even if `--msa-mode single_sequence`

[ajasja]

Expected Behavior

No query should be sent to the server if --msa-mode single_sequence

colabfold_batch --num-recycle 6 --msa-mode single_sequence --model-type AlphaFold2-multimer-v2 --recompile-padding 1 ./{fn} .

Current Behavior

Queries are throttled!

2022-04-14 16:47:58,472 Running colabfold 1.2.0 (6b6898514ec50dfc1a46c2c6cfa787319e7289ff)
2022-04-14 16:47:58,485 Found 2 citations for tools or databases
2022-04-14 16:48:06,026 Query 1/1: nLuc-B_BC-cLuc (length 114)
2022-04-14 16:48:08,204 Sleeping for 6s. Reason: RATELIMIT
2022-04-14 16:48:15,699 Sleeping for 7s. Reason: RATELIMIT
2022-04-14 16:48:25,661 Sleeping for 7s. Reason: RATELIMIT
2022-04-14 16:48:34,795 Sleeping for 6s. Reason: RATELIMIT
2022-04-14 16:48:42,303 Sleeping for 6s. Reason: RATELIMIT
2022-04-14 16:48:50,748 Sleeping for 10s. Reason: RATELIMIT
2022-04-14 16:49:09,517 Sleeping for 5s. Reason: RATELIMIT
2022-04-14 16:49:16,348 Sleeping for 9s. Reason: RATELIMIT
2022-04-14 16:49:26,972 Sleeping for 7s. Reason: RATELIMIT
2022-04-14 16:49:35,504 Sleeping for 10s. Reason: RATELIMIT
2022-04-14 16:49:47,474 Sleeping for 10s. Reason: RATELIMIT
2022-04-14 16:49:59,980 Sleeping for 9s. Reason: RATELIMIT
2022-04-14 16:50:10,974 Sleeping for 6s. Reason: RATELIMIT
2022-04-14 16:50:19,813 Sleeping for 7s. Reason: PENDING

Steps to Reproduce (for bugs)

colabfold_batch --num-recycle 6 --msa-mode single_sequence --model-type AlphaFold2-multimer-v2 --recompile-padding 1 ./{fn} .

Your Environment

{
    "num_queries": 1,
    "use_templates": false,
    "use_amber": false,
    "msa_mode": "single_sequence",
    "model_type": "AlphaFold2-multimer-v2",
    "num_models": 5,
    "num_recycles": 6,
    "model_order": [
        3,
        4,
        5,
        1,
        2
    ],
    "keep_existing_results": true,
    "rank_by": "multimer",
    "pair_mode": "unpaired+paired",
    "host_url": "https://api.colabfold.com",
    "stop_at_score": 100,
    "recompile_padding": 1.0,
    "recompile_all_models": false,
    "commit": "6b6898514ec50dfc1a46c2c6cfa787319e7289ff",
    "version": "1.2.0"
}

[ajasja]

@martin-steinegger I think there should be a case to handle the single sequence mode of paired and unpaired here: https://github.com/sokrypton/ColabFold/blob/f3f924e4d0acc69ebab7083ca895339976e57f12/colabfold/batch.py#L669

[ajasja]

And running with --msa-mode single_sequence --pair-mode unpaired returns

2022-04-14 17:56:17,482 Query 1/1: nLuc-ABC_B-cLuc (length 154)
2022-04-14 17:56:17,486 Could not generate input features nLuc-ABC_B-cLuc: 'NoneType' object is not subscriptable
Traceback (most recent call last):
  File "/home/aljubetic/AF2/CF2/colabfold-conda/lib/python3.7/site-packages/colabfold/batch.py", line 1141, in run
    model_type,
  File "/home/aljubetic/AF2/CF2/colabfold-conda/lib/python3.7/site-packages/colabfold/batch.py", line 839, in generate_input_feature
    paired_msa[sequence_index]
TypeError: 'NoneType' object is not subscriptable
2022-04-14 17:56:17,488 Done

[ajasja]

I'll guess I can get arounds this by making a paired+unpaired a3m my self? (I'm just not sure how a custom a3m alignment is passed down) (@sokrypton)

[martin-steinegger]

I pushed some change that should fix it. Could you give it a try again please?

[ajasja]

Thanks! this fixes the call to the server, but sets paired_a3m_lines = None

Is that the expected behavior. For a single sequence I would expect:

>unpaired
AAAAAAAAAAAAAAAAAAAAAAAAAA-----------------------------------------------
--------------------------BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

>paired
AAAAAAAAAAAAAAAAAAAAAAAAAABBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

>paired+unpaired
AAAAAAAAAAAAAAAAAAAAAAAAAABBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB
AAAAAAAAAAAAAAAAAAAAAAAAAA-----------------------------------------------
--------------------------BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

[ajasja]

@martin-steinegger Sorry I was just running some old code (so you got a lot of traffic from 193.2.14.82). Should be fixed now. ( PS Is [paired_a3m_lines = None] the correct behavior)

[ajasja]

For heterodimers it looks like the coverage is as I would expect: But for homodimers it looks like:

So there is no paired+unpaired.

[ajasja]

@sokrypton Is there any advantage to have multiple copies of the same sequence? I think superfold does something similar.

Exmaple:

unpaired AAAAAAAAAAAAAAAAAAAAAAAAAA----------------------------------------------- AAAAAAAAAAAAAAAAAAAAAAAAAA----------------------------------------------- AAAAAAAAAAAAAAAAAAAAAAAAAA----------------------------------------------- AAAAAAAAAAAAAAAAAAAAAAAAAA----------------------------------------------- AAAAAAAAAAAAAAAAAAAAAAAAAA----------------------------------------------- --------------------------BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB --------------------------BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB --------------------------BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB --------------------------BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB --------------------------BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

paired AAAAAAAAAAAAAAAAAAAAAAAAAABBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB AAAAAAAAAAAAAAAAAAAAAAAAAABBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB AAAAAAAAAAAAAAAAAAAAAAAAAABBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB AAAAAAAAAAAAAAAAAAAAAAAAAABBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB AAAAAAAAAAAAAAAAAAAAAAAAAABBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB

[sokrypton]

For alphafold-multimer putting them on a line, shouldn't be a problem, but for alphafold-ptm... this can be an issue. I'll see if I can fix this in the beta version!

Thanks for the alert.