Structural tokenizer (`PdbQuantizer`) is too slow at processing long proteins

[KatarinaYuan]

Hi Teams,
Thanks for the great work. Wondering how long it takes to process a protein of length 400 for your pre-trained PdbQuantizer? On my machine, somehow it's super slow. Just trying to figure out the reason.

Thanks for help!

[mingchen-li]

Hi, thanks for concerning our work. We will release an accelerated PdbQuantizer with multi-thread parallel processing next month.

[mingchen-li]

Accelerated version speed:

Protein Name (Uniprot_ID)	Length (Local structures)	Splitting to local structure	Encoding
CCDB_ECOLI_Adkar_2012	101	0.29s	4.43s
ESTA_BACSU_Nutschel_2020	212	0.67s	4.27s
PTEN_HUMAN_Matreyek_2021	403	1.06s	4.45s
ENV_HV1B9_DuenasDecamp_2016	853	3.24s	5.63s

[KatarinaYuan]

Thanks! Looking forward to the release!

[SIITW]

Hello, when will it be released?

[mingchen-li]

Hi, please check the new quantizer.py. (Make sure that you have installed pathos in your python env.[pip install pathos])

[GGchen1997]

Still too slow on my side. Could you tell us how to use the new quantizer.py. I just use it as suggested:

from prosst.structure.quantizer import PdbQuantizer
processor = PdbQuantizer(structure_vocab_size=2048) # can be 20, 128, 512, 1024, 2048, 4096
result = processor("example_data/p1.pdb", return_residue_seq=False)

[zz-lovely]

Still too slow on my side. Could you tell us how to use the new quantizer.py. I just use it as suggested:

from prosst.structure.quantizer import PdbQuantizer
processor = PdbQuantizer(structure_vocab_size=2048) # can be 20, 128, 512, 1024, 2048, 4096
result = processor("example_data/p1.pdb", return_residue_seq=False)

I want to know,too! It's super slow. :(

[Linzy19]

it take me about 2 hour to process 30 proteins complex, which 400AA, how can i accelate this progress?

[Tpan1039-ui]

We're excited to share that we've just merged a significant optimization contributed by mdanzi. Which took a batch of 100 proteins from running in about 7 hours to running in about 80 seconds. thanks again to mdanzi for the excellent contribution!