Hello,
I'd like to evaluate the I-TASSER predictions of proteins with no homologous templates (a kind of "hard target") but having known disulfide bonds. Recently, your group published a paper demonstrating how providing atom contacts in the I-TASSER pipeline can greatly enhance the prediction accuracy.
In order to assess the predictive power of I-TASSER in such cases, I have tried to make some predictions of a protein with an already known structure and exhibiting 3 disulfide bonds (1WY6:A from PDB). Here are the following contact restraints that I provided:
DIST 42 CA 69 CA 8.0
DIST 50 CA 78 CA 8.0
DIST 147 CA 155 CA 8.0
CONTACT 42 69
CONTACT 50 78
CONTACT 147 155
In order to convert this protein as a "hard target", I also used these exclusion constraints:
1wy6:A 25
Here are the results I got:
http://zhanglab.ccmb.med.umich.edu/I-TASSER/output/S83825
None of the 5 models can be considered as satisfying, since the RMSD values with the native structure are quite high (9-11A).
I would like to know if you have any recommendations to enhance the prediction accuracy to acceptable levels, or if you feel that no improvements could be made to significantly improve the prediction? Should I also provide the contacts between the cysteines side chains (If so, should I use the SG or the side chain center of mass in the constraint file?)?
Do you think that a similarity of 25% in the exclusion file is too stringent? I set this value based on how you define a protein as a "hard target", and assumed that even if the threshold is low, it won't decrease a lot the number of templates.
Thanks for yours answer,
Julien