Server Dry Run

CHANGES relative to CASP6

1. You will be receiving CASP7 queries from the distribution server in Davis and the results will be collected directly by the Prediction Center (last time there was an intermediate server at the Columbia University). This change requires no extra work from the server curators and we are mentioning it to keep you informed about the details of the experiment.

2. Incorporating suggestions raised at CASP6 meeting and to reduce email traffic between the Prediction Center and predictors, we will be sending emails to predictors in reply to their submission ONLY if the prediction has been rejected or accepted with warning. In this case you will receive an email containing a summary of errors. As before, you will be able to check the submission status of your prediction through the Server Submission Status Page or through Model Viewer facility.

DRY RUN

We will be testing connectivity with registered CASP7 servers in a week, on April 26. To test your server, we will send you an old CASP6 target (T0281). Please respond as you would for CASP7 targets. Of course, these predictions will not form part of the new experiment.

Volker Eyrich is assisting us in transfering CUBIC-based CASP6 server infrastructure to Davis. So, if your server participated in CASP6, we expect the test to be just a formal recheck. For new servers: similarly to CASP6, we will be sending 3 variables to your server's submission URL: the SEQUENCE, the TARGET-NAME and the REPLY-E-MAIL (where to return the results). We will gather the corresponding parameter names from your server's .cgi script. But if your want to specify them directly or notify us about some additional parameters - please send an email to casp@predictioncenter.org.

IMPORTANT THINGS TO REMEMBER

1. Target distribution procedure for the servers.
At the end of each week you will be notified how many targets we plan to release next week. Targets are planned to be released on business days only (Fridays will be used for release just occasionally in case of target overflow only). Requests to all servers will be sent around 9am PST. The submission engine will resend the query if it encounters obvious connecting problems (like network timeouts, 'no response' etc). Failures that go beyond that require special attention, but we'll make every effort to get predictions from servers within 48 hours, as well as we'll make every effort to notify server curators ASAP if we suspect something is not working. The facility that allows checking data flow status by server predictors will be available.

2. Server submission format.
Please submit predictions in CASP format. Your submissions will be immediately verified as submitted by you. CASP warning and error messages will return directly and immediately to you, and will refer to the material you submitted, allowing you to respond in close to a server time frame, if necessary. Note that, as in previous experiments, no prediction will be finally accepted by CASP until it passes CASP format verification.

Below we summarize differences in format of server submissions from that of human groups.

2.1 Header AUTHOR should contain your group name instead of the 12-digit predictor code, e.g.
AUTHOR 3D-JIGSAW
or
REMARK AUTHOR 3D-JIGSAW

2.2. Header SCORE in your prediction is allowable and optional (a real number reflecting self-estimation of the prediction quality).

2.3. For servers, it is allowed to have several models for the same target in one file. If your server returns only one file with several models in it, we will preprocess such a reply. You may add the only one set of required header fields (PFRMAT, TARGET and AUTHOR) at the beginning of the file. Our script will split a multiple models file into separate files (one model per file) and only then we will send each model separately to our verification software (models 6 and higher will be ignored). A PIN number will be assigned to each model separately. Error messages (if any) will be sent for each model separately. So you will be able to correct models (if needed) separately.

3. The timeframe for server prediction acceptance is the following. All server groups have 48 hours since a target posting at the CASP7 web site to return the prediction. A PIN number will be assigned to each model in the submission. Error notification will be sent to the server curator's email address. If your prediction doesn't pass the CASP verification system, you will have additional 24 hours to correct your format manually and resubmit the rejected prediction. To do this, you will need to include the following line in your revised prediction:
REMARK Corrected server submission: PIN_number
and send it by email to: servcorr@predictioncenter.org.
After 72 hours no server predictions will be accepted or altered in any way. If you expect your server to work on prediction generation more than 48 hours, then we would advise you to register a regular (not server) prediction group for CASP7 instead.

If you have any questions, please send them immediately to us at casp@predictioncenter.org .

[Home] Protein Structure Prediction Center