Utilizing Precomputed MSA with Constraints for Complex Structure Prediction

[kehan777]

Hello,
I am working on complex structure prediction using Protenix and have precomputed MSA results.
I would like to incorporate additional constraints (e.g., distance restraints) into the prediction pipeline.
However, I am unsure how to properly integrate these constraints with the existing MSA data.
Could you provide more details or documentation on how to:
Use precomputed MSA results and constraints json in Protenix.
Thank you for your help!

[zhangyuxuann]

@kehan777 you can refer to example_constraint.json and example.json with msa. Add the precomputed MSA in "proteinChain".

[kehan777]

Thank you for your answer, i use example_constraint.json like:
[
{
"sequences": [
{
"proteinChain": {
"sequence": "EVMLVESGGGLVKPGGSLKLSCAASGFTFSNYAMSWVRQTPEKRLEWVAAISGNEGTYTYYPDSVRGRFTISRDNARNNLYLQISSLRSEDTALYYCARYGLVGALDFWGQGASVTVSSASTKGPSVFPLAPSSKSTSGGTAALGCLVKDYFPEPVTVSWNSGALTSGVHTFPAVLQSSGLYSLSSVVTVPSSSLGTQTYICNVNHKPSNTKVDKRVEPKSCDKTHHHHHH",
"count": 1,
"msa":{
"search_tool": "jackhmmer",
"pairing_db": "uniprot",
"pairing_db_fpath": "./database/uniprot/uniprot.fasta",
"non_pairing_db_fpath": "./database/mgnify/mgy_clusters_2022_05.fa",
"msa_save_dir": "./searched_msa"
}
}
},
{
"proteinChain": {
"sequence": "DIQMNQSPSTLSASLGDTITITCRASQNIDVWLNWYQQKPGDIPKLLIYEASNLHTGVPSRFSGSGSGTDFTLAISSLQPEDIATYYCLQGQDYPFTFGSGTKLEIKRTVAAPSVFIFPPSDEQLKSGTASVVCLLNNFYPREAKVQWKVDNALQSGNSQESVTEQDSKDSTYSLSSTLTLSKADYEKHKVYACEVTHQGLSSPVTKSFNRGEC",
"count": 1,
"msa":{
"search_tool": "jackhmmer",
"pairing_db": "uniprot",
"pairing_db_fpath": "./database/uniprot/uniprot.fasta",
"non_pairing_db_fpath": "./database/mgnify/mgy_clusters_2022_05.fa",
"msa_save_dir": "./searched_msa"
}
}
},
{
"proteinChain": {
"sequence": "ESSVSLTVPPVVKLENGSSTNVSLTLRPPLNATLVITFEITFRSKNITILELPDEVVVPPGVTNSSFQVTSQNVGQLTVYLHGNHSNQTGPRIRFLVIRSSAISIINQVIGWIYFVAWSISFYPQVIMNWRRKSVIGLSFDFVALNLTGFVAYSVFNIGLLWVPYIKEQFLLKYPNGVNPVNSNDVFFSLHAVVLTLIIIVQCCLYERGGQRVSWPAIGFLVLAWLFAFVTMIVAAVGVITWLQFLFCFSYIKLAVTLVKYFPQAYMKFYYKSTEGWSIGNVLLDFTGGSFSLLQMFLQSYNNDQWTLIFGDPTKFGLGVFSIVFDVVFFIQHFCLYRKRPGLQAARTGSGSRLRQDWAPSLQPKALPQTTSVSASSLKGDYKDDDDK",
"count": 1,
"msa":{
"search_tool": "jackhmmer",
"pairing_db": "uniprot",
"pairing_db_fpath": "./database/uniprot/uniprot.fasta",
"non_pairing_db_fpath": "./database/mgnify/mgy_clusters_2022_05.fa",
"msa_save_dir": "./searched_msa"
} }
}
],
"modelSeeds": [],
"name": "8dkw_contact",
"constraint": {
"contact": [
{
"entity1": 2,
"copy1": 1,
"position1": 96,
"entity2": 3,
"copy2": 1,
"position2": 28,
"max_distance": 15
}
]
}
}
]

secondly, i modified position2
"name": "8dkw_contact2",
"constraint": {
"contact": [
{
"entity1": 2,
"copy1": 1,
"position1": 96,
"entity2": 3,
"copy2": 1,
"position2": 270,
"max_distance": 12
}
]

However, the position change did not yield significantly different results.

8dkw_contact2_seed_101_sample_0.txt is a .cif file

8dkw_contact_seed_101_sample_0.txt
8dkw_contact_seed_101_summary_confidence_sample_0.json

8dkw_contact2_seed_101_sample_0.txt
8dkw_contact2_seed_101_summary_confidence_sample_0.json

[Anfankus]

Hi @kehan777 , thanks you for actively trying out our features. I tried to reproduce your results on my local environment and on Protenix's webserver, but my results show that the two contacts will get significantly different results. I checked your result and it seems that the constraints are not in effect. Maybe you can check if your branch is constraint_esm and if the contact info is printed in the running log.

[kehan777]

Thank you for your answer, and it work!
protenix predict --input examples/example_constraint.json --out_dir ./output_msa_esm --seeds 101 --use_msa_server --use_esm

another quenstions, can you use existing msa in constraint.json to predict? If possible, please provide examples

[Anfankus]

Yes, Proteinx supports precomputed MSA and constraints at the same time. Just simply fill field msa and constraint in the json file as you showed above.

If it does not work, I suppose it might because you have run pip install protenix before, and this command would install branch main but constraint_esm by default on you local environment. Maybe you can clone Protenix source code and try inference_demo.sh instead of protenix predict ....

[kehan777]

Thank you!