Helpful Tips

Submission Errors

1). You must upload a file in the text format. All files in other formats are rejected.
2). If you have uploaded a PDB file, then it must have SEQRES records and consistent ATOM records. If you have problems with your PDB file, check if the Chain IDs are matching between your SEQRES and ATOM records, or you have any formatting error where the data has been shifted column-wise. Check with a PDB file downloaded from http://www.rcsb.org/
3). Sometimes you may find submission errors even though it appears that the uploaded file is correct in all respects. In such cases try to upload from an UNIX/LINUX machine. The files you may be uploading from a non UNIX/LINUX machine may have an illegitimate end-of-line character and is read as one single line by the UNIX/LINUX editor. If you do not have access to an UNIX/LINUX machine, you can use Notepad in Windows/Mac to open your UPLOAD file to check if the file is OK.

NO PDB FILES WILL BE ACCEPTED BY PROKNOW UNLESS IT IS INTERPRETABLE BY THE PARSER

Your FSSP files should also be in text format and readable by a text editor.

Interpretation Tips

Scenario 1. You have known structure but unknown function.
First check if your structure has any domain boundaries. If you can visually detect domain boundaries, dissect the protein into the domain parts and submit them to DALI (http://www.ebi.ac.uk/dali). Upload all the results for individual domains into this server. If you cannot detect any domains arbitrarily dissect the protein into meaningful smaller parts and submit your jobs on individual parts. This way you can get different predictions and see which one is the best evidenced. In all these cases please make sure that your dissected protein looks like a normal PDB file with SEQRES records and the ATOM coordinates.

Scenario 2. You have known structure and unknown function and novel fold.
If your protein has a novel fold, then fold information is not useful in detecting any function. Extract the sequence of your protein and submit it to the server. The server will do a fold recognition on your sequence and then return function predictions. Go to the "Click here to get a description of your clues" and look at the PDB files used to predict your function.Manually check for similarities between your file and the PDB files used to make the prediction (For example: check the 3D motifs present, topology of the protein, surface features etc.). If your PDB file has some similarities with the PDB files used for prediction, you can assume the prediction to be more reliable.
Alternately, if you are aware of proteins whose fold is similar in parts to your protein, run DALI on those proteins (PDB files) and upload the DALI (http://www.ebi.ac.uk/dali) output files into this server for job submission.

Scenario 3. You have known structure and known function, but suspect that your protein does something more.
The server will usually return more functions than already known for the protein. These functions will be at a lower Bayesian Weight. You can do biochemical experiments to check for these results. You could also break up the protein and submit jobs with dissected proteins and check the results obtained. In all these cases please make sure that your dissected protein looks like a normal PDB file with SEQRES records and the ATOM coordinates.
Alternately, submit only sequence and see what functions are predicted using fold recognition and analyze the PDB files used for prediction to come to a conclusion on the predicted functions.

Scenario 4. You have unknown structure and unknown function, and have some information on the domain boundaries.
Submit the total sequence and the individual domain sequences for jobs to the server. Check the evidences by clicking "Click here to get a description of your clues" to find consistency in the predicted results for individual domains and the whole protein.

Scenario 5. You have unknown structure and unknown function, and no other information.
Submit the whole protein and also randomly dissected parts and check for the evidences ("Click here to get a description of your clues") to come to your decision. You may get protein information on folds, motifs etc., whic may help you to come to intuitive conclusion. Prior knowledge from scientific literature can be very useful for this purpose.

Scenario 6. I have gotten results but they are vague: like catalysis, transporter etc.
Check the "Click here to get a description of your clues" and see when kind of proteins/motifs/folds etc. are used to predict the function. This can help in infering the results in a way that is less vague.

Scenario 7. I have gotten too many predictions and can't make a sense out of it.
Get the predictions you are interested in, and plot them on to the Gene Ontology Directed Acyclic Graph. For plotting, click on the radio button under "Networked information on ORFs and Ontologies" and besides "Get map connecting Ontologies". Enter the GO codes separated by comma in the box aligned against the button. You may enter any number of GO codes. Choose Depth threshold "Include only parent" for a less cluttered graph. You may change it in anyway you want. Also change the page size and image size as per your requirement. Here the picture will tell you how the different predicted functions are interrelated to each other. Use the predicted Bayesian weights and your judgement to analyze how the functions are clustered in the graph. You could transfer the weights among the predicted functions if they share parent-child relationships. Since the Gene Ontology Graph is non hierarchical, there is no best way to port Bayesian Weights among parent and child terms. Use your personal judgement based on the plot to come to a decision.

Secenario 8. Forget about interpreting results! I have not been able to get results from the server.
1). Add SEQRES records to the file, in case they are not present.
2). Submit your model to DALI (not necessary if it is a novel fold).
Upload the text DALI output file to the ProKnow server.
3). Go to the NCBI site and get the species code for the organism to which your protein belongs. For example: Species code for Mus musculus is 10090 and Arabidopsis thaliana is 3702.
If you get a blank email, go to the Known Problems page to double check your submission. Send me an email if you need help.