3D-JIGSAW is an automated system to build three-dimensional models for proteins based on homologues of known structure.
Further information on Structure Prediction can be obtained from this Guide by Rob Russell.

Submission instructions
Go to submission and paste a protein sequence in one letter code, filling in your e-mail address and a short protein identifier.
Then choose a building mode (automatic or interactive) and submit it.

Automatic mode
The program looks for homologous templates in our sequence databases (PFAM+PDB+nr) and splits the query sequence into domains. If good templates are found, the best covered domain is then modelled using a maximum of 2. This process can take up to an hour, depending on the load of the system. You will receive an e-mail with the alignment between query and template/s and a PDB formatted set of coordinates, that you can display easily with Rasmol.

Interactive mode
The program looks for homologous templates in our sequence databases (PFAM+PDB+nr) and splits the query sequence into domains. An e-mail is sent back to you with a link to a graphical display of this domain arrangement and useful information extracted from the PFAM database. From this link you can choose the domains you need to model and you can select the templates and correct the alignments before submitting a modelling job. Templates are ranked according to the coverage of the query, their sequence identity and their crystallographic resolution. Information from each template is easily accessed, including its alignment to the query sequence:

namefromtoexplanation
1e0x_B
Model! 		1308<- this is a global template, spanning more than one domain
1fh9_A
Model! 		25250<- this is a local template, spanning one domain only

Submit your query and, like in the automatic mode, the results will be sent to you by e-mail.

Even if there are no good templates to model your sequence, you can sometimes recognise the fold of some domains because the corresponding PFAM families may contain proteins of known structure.

How to display the models
Models are sent inside a plain text e-mail. Just save this e-mail as a text file and open it with Rasmol or any other program to display proteins. The model coordinates are written in PDB format, the most widely used for proteins.

Accuracy of the models
Chothia and Lesk showed in 1986 how homologues share their folds depending on their sequence identity. The idea is: the higher the sequence identity between two proteins, the better agreement in their structural superposition. They proposed a formula to calculate the expected average deviation per backbone atom from the overall sequence identity:
deviation [angstroms] = 0.4 exp (1.87 x sequence difference)
Apart from this natural limitation, modelling methods introduce more errors. In general, the less reliable parts of a model are the exposed loops, while secondary structure elements tend to be conserved.