Question about model reliability

[bmlane]

I submitted two jobs in early June on i-Tasser, one for a canonical protein sequence (S739254) and one for a single amino acid substitution (S740311). I saved some information from each job including images for the top model prediction for each job but neglected to save the tarball file for either. When I realized my error, I resubmitted the job requests but received very different model predictions (S744449 and S745092 respectively). I am fairly certain that I submitted the same sequence for both sets (June vs July) but can't access the June submission files to double check. Any explanation for the different outcomes or suggestions for which models to use? I submitted the canonical sequence again for i-tasser-mtd but got a third different model prediction (ITM692624992). Any suggestions? Thanks for the help

[jlspzw]

Dear user,

It is possible that two different runs of the I-TASSER method give you two different models, I-TASSER depends on the template library, and the library is kept updating so the results may differ, especially for those "hard" targets that model has low predicted confidence score. I-TASSER-MTD is designed for handling multi-domain proteins based on the C-I-TASSER algorithm, so it differs from I-TASSER. C-I-TASSER is better than I-TASSER in general behaviorar.
And you may also try our most state-of-the-art method, D-I-TASSER, https://zhanggroup.org/D-I-TASSER/

Best
IT Team

[bmlane]

Thank for the information, very useful going forward. Unfortunately, my protein of interest is just over 1,000 residues, so I do not think C or D-I-Tasser will work for me. Also, is there any way to recover the tarball or model 1 pdb files for the expired jobs (S739254) and (S740311)? Thanks again

[jlspzw]

Dear user,

Your results seem to have been auto-deleted. Unfortunately, we can not recovery them.

Thank you

Best
IT Team