Jpred Old News
10 May 2007
- Fixed a bug in launching the Jalview applet on predictions done on 'msf' format files.
- Updated Jalview to the latest version (v2.3).
13 April 2007
- We have recently had to set limits on webservice jobs due to excessive memory usage by some jobs. However, it seems that the limits were a little too stringent and many Jpred jobs have failed as a result. We have now relaxed the limits and the vast majority of jobs will complete normally. Please let us know if you are still having problems.
10 April 2007 -IMPORTANT
- UPDATE - the migration has been pushed back owing to external factors. We will let you know asap when it is due to happen. Jpred will be running as normal until then.
- We are planning a major upgrade to the Jpred server machine, which will entail migrating Jpred to a temporary server and then back again once the upgrade is completed. This should cause minimal downtime.
Assuming a test migration completes successfully we will move Jpred on Friday 13th April.
12 February 2007
- On the results page a link to the pdf version of the output has been added. This should print better than the postscript version.
22 January 2007
- Over the weekend we had a filesystem problem with the cluster, which caused Jpred jobs to be queued until the error was fixed. The error was fixed this afternoon and all jobs should have completed as normal.
12 December 2006 -IMPORTANT-
- UPDATE 14-12-2006 - The maintainence was completed successfully and Jpred is up again from 1400 GMT.
- Following on from Monday's failed update Jpred will now be done again from Wednesday 13th December at 1700 GMT. Hopefully this time it will be competed successfully and Jpred will up and running from 14th December approx. 0900 for all of your predictions over the Christmas period ;)
08 December 2006 -IMPORTANT-
- UPDATE - Unfortunately the update was not completed over night. So will be done at a later time. Jpred will have to come down again some time soon.
- Please note that Jpred is going to go down from 1700 GMT on Monday 11th December. All submissions after that time will be rejected. This is due to campus-wide network changes, but should be back on Tuesday morning.
27 November 2006
- The Jpred server had to go down at short notice to resolve technical issues with our network. It was down for less than two hours and now all is working as normal.
4 October 2006
- We have been having a few problems with job submissions over the last few days - with some jobs randomly (it seems) failing to be submitted to the queue. We are aware of the problem and will solve it as soon as possible. Apologies for any disruptions.
22 September 2006
- Fixed a couple of bugs:
1. The queue check link now works properly.
2. If a Jpred job fails the appropriate error is emailed to the user.
22 August 2006 - IMPORTANT -
- The Jpred server is going down at close of play (approx. 17:00 GMT+1) on Thursday 24th August. All current jobs will continue until finished, but new submissions will be rejected. Hopefully, it will be back in service on Monday 28th August. Apologies for any inconvenience casued.
05 June 2006
- As you may have noticed we have been having a few techinical problems over the last few days. Today the server has been brought down to try and rectify things. It will be back up as soon as possible.
23 May 2006
- The server has been moved to a new cluster. All the code remains essentially the same, but the
odd bug may have crept in. Please let us know if you spot any dead links or errors at email@example.com
- The PDB and Swissprot-TrEMBL (now Uniprot) databases have been updated to the latest versions.
This is a significant update as Uniprot has grown from 0.5M to >3M sequences. This will improve accuracy, but speed
will be affected. Again, let us know if this causes problems.
22 May 2006
- A bug in generating the postscript and pdf outputs has been fixed.
18 May 2006
- The JalView applet used on the results page has been updated to the latest version. It now recognises JNet output and deals with it accordingly.
16 Feburary 2005
- The server has had problems with jobs that provided an email address, which was causing it to lose jobs once they had finished.
22 November 2004
- One of our servers suffered a meltdown over the weekend, which caused jobs not to run. This problem has been fixed now. If you suffered from this, apologies.
- Some users may have been experiencing errors with jobs, where everything is complete, but the final results page is not being generated. This should now be fixed.
- Because of the demand that the jpred service is getting, we have to reduce the length of time that jobs are stored for to two days and the PSIBLAST searcg results are now compressed.
- Jpred has had many changes applied, hopefully making it even more stable.
- The server has been benchmarked against the same test set in the Cuff & Barton (2000) paper, and has the same accuracy.
- Due to space considerations, results are now only stored for 3 days.
- Jpred adapted to run on our Linux cluster, bug fixes applied.
- Webside updated.
- Jpred now consists only of Jnet. This is due to problems with portability of the other prediction methods across operating systems.
- Jpred still uses a 3 iteration PSI-BLAST search to generate an alignment and profile to make a prediction from.
- Jpred still uses a filtered SWALL database to do the PSI-BLAST.
- ClustalW is no longer used to create the alignment.
Jpred at the EBI
- Jpred now includes Jnet, a neural network secondary structure prediction method (see here for more information about Jnet). Jnet is the default prediction mode.
- Jpred includes version 1.2 of DSC, this method has an updated and more accurate GOR matrix, which was found by the authors to improve prediction accuracy by 0.5 to 1%.
- Jpred uses a 3 iteration PSI-BLAST search to obtain sequences for secondary structure prediction. The profiles generated by PSI-BLAST are vital for the Jnet secondary structure prediction method.
- A filtered sequence database is used to search for related proteins to the query sequence. The filtering masks any low complexity, tramsmembrane, or coiled coil regions. The software for filtering the database was written and kindly supplied by David Jones. Seg and Coils are then also applied to the sequence database to mask any regions missed by HELIXFILT.
- Using a filtered database to generate alignments for secondary structure prediction is vital for the Jnet method, Jnet's accuracy improves by 1% with a 7 fold cross validated test using the filtered database. However, the older methods (PHD, DSC, PREDATOR and NNSSP) are no more accurate using this method, but are no worse either
- Jpred now predicts coiled coil regions with COILS and MULTICOIL within your protein
- PHDhtm predictions are also now displayed
- A parallel version of CLUSTALW from SGI applied, to speed up the alignment building process
- The MAP interface has been removed. Development of a new version of MAP is currently underway in our group
Even Older news
- November 1999: Jpred2 is born!: Bigger, better, faster secondary structure predictions :-)
- August 1999: Jpred has done over 5800 predictions
- Jun 30 1999: Jpredusers mailing list is now active
- Jun 1999: Jpred does 1249 predictions this month
- Apr 28 1999: Jpred is getting busy, we've put in a queue checking device so you can see how many jobs are waiting
- Mar 1999: 547 Jpred predictions this month!
- Mon Mar 1 1999: Proteins paper published, links to distribution data included, and references
- Dec 1998: Bioinformatics paper published
- Fri Nov 13 1998: OWL database replaced with SWALL
- July 1998: CASP3 Target predictions
- Sun Oct 25 1998: An experimental interface to MAP has been included
- Sun Oct 25 1998: Jpred now searches known structures for sequence similar proteins before prediction can be carried out