search

RosettaCommons / RFDesign

Protein hallucination and inpainting with RoseTTAFold
Other
239 stars 66 forks source link

about inpainting details #2

Open guyujun opened 2 years ago

guyujun commented 2 years ago

Hi, developers

i am trying to play the inpainting method, i use the test case 2KL8.pdb，and i keep the contigs from A2-8 and to rebuild the left structure.

what i use the command:

!python inpaint.py --task hal --checkpoint v02 \ --pdb=./test/input/2KL8.pdb \ --out=./test/output/inpaint_test2 \ --contigs A2-8 --len 100 \ --num_designs 1 \ --inf_method=multi_shot --n_iters 10 --recycle_str_mode both --clamp_1d 0.1 --dump_all

result: Generating design 0 (1 total)... mask_str (1=rebuild; 0=keep): 1111111111111111111111111111111111111111111100000001111111111111111111111111111111111111111111111111 mask_seq (1=rebuild; 0=keep): 1111111111111111111111111111111111111111111100000001111111111111111111111111111111111111111111111111 0 LLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLEMDIRFRLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL 1 KLLLLLLLLKELLLKLLLKLLLKLLKKLKLLLLELLELLLKLLEEMDIRFRLLLLLLLELLLELLEELLELLLLLLKKLLLLLLLLLLLLLLLLLLLLLL 2 SLKLEESLLLLELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLEMDIRFRLNLELLEELLELLLELLLLLLELLKLLLLLLLLLLLLLLLLLLLLLLLL 3 KLKLEELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLEMDIRFRLLLELLEELLELLLELLLLLLELLLELLLLLLLLLLLLLLLLLLLLLLL 4 SLLLKLKLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLKLLLEMDIRFRLLLELLLELLELLLELLLLLLLLLLKLLLLLLLLLLLLLLLLLLLLLLL 5 KLKLPESLLLLLLLLLLLLLLLLLLLKLKLSLLLLKLLLELLLKEMDIRFRLSGELLEELLELLLELLELLLELLLLLLLLLLLLLLLLLLLLLLLLLLL 6 SLKLLELLLLLLLLLLLLLLLLLLLLGLKLSLLLLLKLLELLLEEMDIRFRLLLGLLLLLLELLLELLELLLELLLLLLLLLLLLLLLLLLLLLLLLLLL 7 SLKLLELLLLLELLLLLLLLLLLLLLGLKLSLLNLKKLLKLLLEEMDIRFRLLLGLNLELLELLLLLLLLLLELLLLLLLLLLLLLLLLLLLLLLLLLLL 8 SLKLLELSLLLELLLLLLLLLLLLLLLLKLSLLNLLELLELLLKEMDIRFRLLLGLNLELLELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL 9 SLKLLELSLLLELLLLLLLLLLLLLLKLKLSLLLLLELLKLLLKEMDIRFRLLLGPLLELLELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL Designed sequence (excluding receptor): SLKLLELSLLLELLLLLLLLLLLLLLKLKLSLLLLLELLKLLLKEMDIRFRLLLGPLLELLELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL Saving outputs for /content/RFDesign/inpainting/test/output/inpaint_test2_0: pdb npz fas trb Finished design in 7.80 seconds. Max CUDA memory: 2.1457G

after 10 iters, the sequence seem abnormal, can you giving me some suggestions to make it better？

thanks, weikun

jueseph commented 2 years ago

This output is expected. These inpainting checkpoints aren't good at building regions longer than ~20 residues.

If you, for instance, use --window A,2,8 (rebuild that region only), you'll get more sensible results.

If you want to scaffold a motif into a larger protein, you should use hallucination. We currently only use inpainting to diversify small regions of structure & sequence in a protein that has already been designed using other methods. We are currently training some networks that are better at longer-region inpainting; those will probably be released with the published version of the paper.

On Sat, Jan 15, 2022 at 8:20 PM guyu @.***> wrote:

Hi, developers

i am trying to play the inpainting method, i use the test case 2KL8.pdb，and i keep the contigs from A2-8 and to rebuild the left structure.

what i use the command:

!python inpaint.py --task hal --checkpoint v02 --pdb=./test/input/2KL8.pdb --out=./test/output/inpaint_test2 --contigs A2-8 --len 100 --num_designs 1 --inf_method=multi_shot --n_iters 10 --recycle_str_mode both --clamp_1d 0.1 --dump_all

result: Generating design 0 (1 total)... mask_str (1=rebuild; 0=keep):

1111111111111111111111111111111111111111111100000001111111111111111111111111111111111111111111111111 mask_seq (1=rebuild; 0=keep):

1111111111111111111111111111111111111111111100000001111111111111111111111111111111111111111111111111 0 LLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLEMDIRFRLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL 1 KLLLLLLLLKELLLKLLLKLLLKLLKKLKLLLLELLELLLKLLEEMDIRFRLLLLLLLELLLELLEELLELLLLLLKKLLLLLLLLLLLLLLLLLLLLLL 2 SLKLEESLLLLELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLEMDIRFRLNLELLEELLELLLELLLLLLELLKLLLLLLLLLLLLLLLLLLLLLLLL 3 KLKLEELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLEMDIRFRLLLELLEELLELLLELLLLLLELLLELLLLLLLLLLLLLLLLLLLLLLL 4 SLLLKLKLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLKLLLEMDIRFRLLLELLLELLELLLELLLLLLLLLLKLLLLLLLLLLLLLLLLLLLLLLL 5 KLKLPESLLLLLLLLLLLLLLLLLLLKLKLSLLLLKLLLELLLKEMDIRFRLSGELLEELLELLLELLELLLELLLLLLLLLLLLLLLLLLLLLLLLLLL 6 SLKLLELLLLLLLLLLLLLLLLLLLLGLKLSLLLLLKLLELLLEEMDIRFRLLLGLLLLLLELLLELLELLLELLLLLLLLLLLLLLLLLLLLLLLLLLL 7 SLKLLELLLLLELLLLLLLLLLLLLLGLKLSLLNLKKLLKLLLEEMDIRFRLLLGLNLELLELLLLLLLLLLELLLLLLLLLLLLLLLLLLLLLLLLLLL 8 SLKLLELSLLLELLLLLLLLLLLLLLLLKLSLLNLLELLELLLKEMDIRFRLLLGLNLELLELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL 9 SLKLLELSLLLELLLLLLLLLLLLLLKLKLSLLLLLELLKLLLKEMDIRFRLLLGPLLELLELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL Designed sequence (excluding receptor):

SLKLLELSLLLELLLLLLLLLLLLLLKLKLSLLLLLELLKLLLKEMDIRFRLLLGPLLELLELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL Saving outputs for /content/RFDesign/inpainting/test/output/inpaint_test2_0: pdb npz fas trb Finished design in 7.80 seconds. Max CUDA memory: 2.1457G

after 10 iters, the sequence seem abnormal, can you giving me some suggestions to make it better？

thanks, weikun

— Reply to this email directly, view it on GitHub https://github.com/RosettaCommons/RFDesign/issues/2, or unsubscribe https://github.com/notifications/unsubscribe-auth/AABCU53YWADGFBG6IUTD5ILUWJBPTANCNFSM5MBZWXIA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

You are receiving this because you are subscribed to this thread.Message ID: @.***>

guyujun commented 2 years ago

it is very nice of you, thank you for the details!