Disclaimer and Terms of Use
Visitors To Date:
Welcome to the 3D-PSSM Web Server V 2.6.0

A Fast, Web-based Method for Protein Fold Recognition using 1D and 3D Sequence Profiles coupled with Secondary Structure and Solvation Potential Information.

The successor to 3dpssm is ready for use (beta-testing)
The new system is known as Phyre and has an up-to-date fold library, 10-15% better coverage than 3d-pssm, and a new interface.

Please Note: 3D-PSSM is now for academic use only.

3D-PSSM frozen - new system shortly Jun 18 2004
I have today frozen the updates for 3D-PSSM. New pdb structures will no longer be added to the fold database.
I intend to release our new server to the public in September. Benchmarking has shown the new system to be a substantial improvement over 3D-PSSM.

Lawrence Kelley
Completely new protein structure prediction system Apr 5 2004
A brand new fold recognition system is on its way. The new system is nearing completion. Benchmarking suggests it is far superior to 3D-PSSM. However, it requires substantially more CPU power. As a result I will probably have to shut down 3D-PSSM once the new system is up and running. More details to follow.

Lawrence Kelley
Temporary removal of d3btaa3 Mar 26 2002
The domain d3btaa3 is problematic and its sequence profile has drifted out of control. I have temporarily removed this template from the library until I can fix it. It can give strong false positive hits because its profile has become very flat (generic) matching many different and unrelated sequences.

Move to Imperial College of Science, Technology and Medicine Mar 4 2002
The 3DPSSM service has moved to Imperial College, London. There may be a few teething troubles with the new system at first, so please let me know of any problems you experience with the system. E-mail me, Lawrence Kelley at L.A.Kelley@ic.ac.uk

3D-Genomics Nov 13 2001
Protein structure and domain assignments for fully sequenced genomes are available from our 3D-GENOMICS database

Fold Library Now SCOP 1.53 Aug 1 2001
The fold library is now 'up-to-date' with SCOP 1.53. As I was updating the library SCOP decided to move to version 1.55 with a whole bunch of format changes. So for the time being 1.53 will have to do. Many people's results may be different now compared to the older 1.50 library. Everything is currently in a beta phase, and I would like to hear from anyone who gets strikingly different results now compared to a few days ago, or whether there are any problems with the new library. (Lawrence Kelley)

Fold Library SCOP 1.53 Update and Browser Downtime Jun 19 2001
I have begun the monumental task of updating the fold library to be in sync with SCOP 1.53. Most things should work fine (famous last words) while the update is in progress EXCEPT FOR THE FOLD LIBRARY BROWSER. The library browser will be unavailable until the update is complete, which could be for the next several days. I will make an announcement when it's back up.

If you have any difficulties or the system appears to be behaving strangely please let me know (Lawrence Kelley)

Biotext (and a bugfix) May 9 2001
I have incorporated some work going on in the lab by Ben Stapley called Biotext. This is an enhanced version of the SAWTED textual annotation system. In the experimental version I have incorporated into the server, your query protein is represented by a set of SwissProt Keywords taken from Swissprot homologues to your query. In addition, similarly to SAWTED, a cosine similarity between said keywords and corresponding keywords in each of the top 20 structural hits is presented. Contrary to SAWTED, scores closer to 1 are good, and 0 is bad. This will probably change in the near future.

Also a slight bug fix: The text e-mail e-values were wrong and did not correspond to the HTML web pages. This unfortunately has been the case for a few weeks without me noticing! However, this is now fixed.

Server Problem (Fixed) Apr 19 2001
Apologies to users who have submitted sequences this morning. There was a problem with the automatic library updating, which I have now fixed. However, I'm still testing the update script and there may be a few more problems on Wednesday night/Thursday morning next week when it updates again.

A variety of small changes Apr 14 2001
Just a bunch of small changes made over the last couple of weeks:
  • Pairwise sructural alignments are finally visible in the fold library browser on PCs using CHIME. (This problem has been around since the browser was made available but I've finally found a workaround).
  • You may now choose to do only 3 iterations of PSI-Blast (instead of the default 5) on the advanced submission page. This may be helpful if your sequence has drifted. Check out the help page for more info.
  • For multiple sequence alignment submissions, I've made it so your alignment gets converted by CLUSTALW itself. If your multiple sequence alignment can be understood by CLUSTALW, then it should be understood by the server. This means multiple sequence alignments in FASTA format *should* work ok.
  • A binary, memory-mapped version of the fold library is now used by 3dpssm, and as a result jobs only take about 2-4 minutes of 3dpssm time. The rest of the time for your job will be due to PSI-Blast,scwrl, and other peripheral programs. Overall, you should see an increase in turnaround time, although occasionally SCWRL can take a little while.
  • I've altered the way SAWTED is implemented slightly. Now, if you get a 95% confident hit or better at rank 1, SAWTED scores for everything else are ignored. This doesn't mean you should ignore it. It just means we avoid some silly cases where high %id hits come lower down than remote homologues due to confident SAWTED hits. However, this low-ranking of high %id hits can still sometimes happen due to 5 iterations of PSI-Blast moving into remote homologue territory. I'm currently looking into fixing this.
  • My implementation of SCWRL was slightly buggy and I've now fixed this. Sometimes sidechains wouldn't be added to all the top five confident hits, and sometimes jobs would die from taking too much CPU time. This should now be remedied.
  • People submitting multiple sequence alignments from mac's would often screw up the server due to end of line characters in mac text. This has been rectified.
All these subtle changes combined constitute a slight increase in server version. I am not changing the binary version number. Although the binary is different in technical respects, scientifically its no different. The algorithm is the same - just a different implementation.
P.S. Oh, you'll also notice the little status bar near the top of the page to keep you up-to-date with when the fold library was updated, and how many entries there are. The library is updated weekly in synch with the PDB.

Server Version 2.5.6

New sidechain modelling with SCWRL Mar 25 2001
I've incorporated a new experimental feature into the server. For the top 5 high confident (>95% / red hits) side chains will be modelled for your sequence using Roland Dunbrack's SCWRL program. This may increase the time required for some jobs, but very soon I will be also incorporating a speed-up in 3d-pssm which should easily compensate for any increased processing time caused by SCWRL.

When sidechains have been added to your model, an extra link will appear below the gif image of your structure. Clicking on this will either download the coordinates or launch rasmol, depending on how your browser/system is set up.

Please remember this is experimental.If anything starts going wrong that worked before, please Let Me know. Also, if you find it useful (or not) I'm keen to hear your comments.

Server Version 2.5.5

New PSI-Blast Version 2.1.2 Mar 22 2001
I've upgraded our PSI-Blast version to 2.1.2. This version is supposed to have better statistics than earlier PSI-Blast versions. As a result, some people's results may change if re-run with the new version. As usual, if you get radically different results now compared to previously, please let me know.


Server Version 2.5.4

Experimental New PSI-Blast Parameters Feb 8 2001
I have altered the parameters for the PSI-Blast run that is initially performed on your sequence in the 3D-PSSM server. I've increased the number of iterations performed (to 5) and slightly loosened the h-value (which determines when a sequence is included in the profile for the next iteration [from 0.001 to 0.002]).

This means that if your sequence looked like an orphan (had no homologous sequences) in previous 3D-PSSM runs, this may no longer be the case. That is, the server may now find homologues it missed before, and as a result generate a better secondary structure prediction, and have a higher probability of finding a good structural match in our library.

I'm not sure if I will keep these parameters as it adds extra load to our server, and may bring in false positives due to PSI-Blast drift. I'll see how it goes. Please let me know if you get radically different (good or bad) answers as a result of this change.

More Hardware = Faster Turnaround Jan 11 2001
We have approximately 8 new processors for web jobs now, so the queue should clear much more quickly from now on. As usual, there may be some hiccups along the way, so please E-mail me if things seem to be breaking/not working like they used to. Happy submitting...

Improved Secondary Structure Prediction - PSIpred2 Dec 13 2000
I have integrated version 2 of PSIpred from David Jones (http://insulin.brunel.ac.uk/psipred/) into the server. The improved secondary structure prediction accuracy (approx 80%) should lead to improved recognition of folds and possibly improved alignment accuracy (although this will be dealt with separately in the near future). This has warranted a server version number change. We are now at Server Version 2.5.3.

Simplified submission page and default global-local Dec 9 2000
I have simplified the submission page. There is a link at the top of the submission page to the advanced page where you can still do things like upload your file, submit multiple sequences, switch between global and global-local algorithms, etc.
By default, your submission will now use the global-local option. Tests and the results from CASP4 meeting have indicated that this method is by and large superior to the previous default method - global. This has warranted a server version number change. We are now at Server Version 2.5.2.

Searchable Fold Library and Renewable Results Aug 8 2000
The fold library (see menu to the left) now has a simple search facility permitting you to find folds, superfamilies, families, etc. in our library containing a search term or terms. It is not a full boolean search and order of terms is important, but it should help you find things a bit quicker.

It is now possible to renew your results on the server so they are not automatically deleted after four days. On the HTML results page near the top next to "Download your results" is "Renew your results". Clicking this option will ensure your results are kept on the server for 4 more days from the current date.

CLUSTAL Input and Javascript SAWTED July 14 2000
The server can now accept a multiple sequence alignment in CLUSTAL format. This can be useful in cases where you may have homologous sequences that cannot be found by our 3 iterations of PSI-Blast. For example, you may have searched various DNA databases. Having multiple homologous sequences can radically improve Secondary Structure prediction, and create a more powerful sequence profile (or 1D-PSSM), both of which can improve the likelihood of useful fold recognition results.

Also I have altered the way in which SAWTED keywords are displayed on the results page. On some browsers (well on Mac's) the "alt=blah" img tag parameter didn't work properly, and it was not possible for people to see the keywords. Now the words also appear in the status bar, and will be displayed in a Javascript alert pop-up if clicked on. Hopefully this will cover most eventualities.

Browse and Explore the Fold Library June 6 2000
I'm opening up a major new aspect to the 3D-PSSM server today. Finally, the link in the menubar to the left for "Fold Library" actually goes somewhere! You may browse through a hierarchical tree-structured representation of the library which is essentially a cut-down version of SCOP.

When you find the protein you want, you will be presented by a representation of the sequence,accessibility,secondary structure, and core residues as well as a variety of links. These links permit visualisation of the multiple sequence and multiple structure alignments, in a variety of views (secondary structure,accesibility,etc). In addition, you may view the actual sequence profiles (PSSMs) used by the server. You can also bring up 3D superpositions of structural homologues (using Rasmol).

It is hoped that these facilities will enable users (and myself) to better appreciate the nature of the diversity and similarity on both the sequence and structural levels of potential hits produced by the server. It may also highlight advantages/shortcomings in our library, which may help us improve it in the future. Your feedback is very welcome.

As an example you can jump straight to a Globin protein d3sdha_

JMB Paper reference May 19 2000
The 3D-PSSM paper in the Journal of Molecular Biology describing the method will be published on 2 June 2000. The full reference is:

Kelley LA, MacCallum RM & Sternberg MJE (2000). Enhanced Genome Annotation using Structural Profiles in the Program 3D-PSSM. J. Mol. Biol. 299(2), 501-522

New Fold Library - approx. 5000 templates - automatic updating May 10 2000
The new, bigger, and hopefully better fold library is now in place.

What's New?

Fold Library

  • Based on SCOP 1.50 plus extra PDB structures not yet incorporated into SCOP
  • Enlarged database (almost 5000 entries, previous size 3000).
  • Non-redundant to 70% sequence identity
  • True SCOP domain codes are used (previously domain codes were unique to us).

  • Versions

    I have included in both the html and e-mail results output, some version information for the sequence and template libraries, the server, and the 3dpssm binary.

  • The library version number is composed as:[< SCOP version >. < number of templates >]
  • The sequence database version is simply the date of its last mirroring from the NCBI.
  • 3D-PSSM binary and overall server versions

  • SAWTED Improvements

  • Extra stopwords (meaningless words) are screened and SAWTED now uses SWISS-PROT release 38.0

  • Regular Automatic Updates

    As new structures appear in the PDB, they are checked against the library, and added if necessary. This, along with a system to keep current with SCOP, should mean the system will always be up-to-date. (fingers crossed).

    As this is its first release, it has not yet been extensively tested. So any feedback from you, the users, would be most appreciated. Comments/Complaints/Suggestions to me (Lawrence Kelley).

    New Fold Library very soon Apr 26 2000
    Within a few days (probably this week) the new improved fold library will be incorporated into the 3D-PSSM server. This will increase the number of representative structures from approximately 3000 to 5000. In addition, the library will be automagically updated (every few days or every day - not sure yet) as new PDB structures are released. This, along with a variety of bug-fixes, and possibly a new post-filtering technique, will amount to a new version of 3D-PSSM (Version 2.5). Also, version information will be returned with each submission, and the versions will change with each update of the library (2.5.0,2.5.1,etc).

    In the slightly more distant future, the fold library and its accompanying derived information will be browsable through the web (finally).
    Paper in Journal of Molecular Biology Apr 6 2000
    A paper describing the 3D-PSSM methodology is in press and will be available shortly. This paper will be the primary citation in the future.

    Kelley LA, MacCallum RM & Sternberg MJE (2000). Enhanced Genome Annotation using Structural Profiles in the Program 3D-PSSM. J. Mol. Biol. in press

    Multiple Sequence View Mar 28 2000
    I have altered the viewing mechanism (i.e. the mview output parameters) for the multiple sequence alignment of your query. The previous system, although looking nice, could not be properly handled by MS Internet Explorer, but could be handled fine by Netscape Navigator. I have opted for the lowest common denominator so everyone can easily view their multiple sequence alignments.

    Orphan Rescued Mar 2000
    Managed to fix the orphan problem. Wish someone had alerted me to this earlier :(. Nevertheless, all is now right in the world of 3D-PSSM and its orphans.

    Orphan Problems Mar 2000
    The server is currently not processing orphan sequences at the moment. When PSI-Blast is run on an orphan (i.e. a sequence with no sequence homologues) no PSSM is generated and the fold recognition process dies. I will fix this a.s.a.p. Apologies for any inconvenience caused. I will make an announcement when it is fixed.

    PROSITE Added Feb 25 2000
    I've added a PROSITE search to the server. This is run on your sequence and results are displayed in the HTML output. If anything is found, you can click on the PS accession number to get more details (from www.expasy.ch). Although it will probably be rare to get a PROSITE match, I thought it would be useful just in case something obvious is missed by the server. In the future I intend to mark up the locations of prosite motifs in the library structures in the alignment view.

    Help page Added Feb 24 2000
    After a very long wait I have finally done version 1 of the help page. Click on the help option in the menu bar on the left to have a look. Hopefully this will address some of the questions people have had. If I have missed something out which you want to know, or something is not clear in the help, then please let me know and I can do something about it.

    Bug fixes and optional Seg Feb 16 2000
    Found a bug in the global-local option (oops) which is now fixed. It probably wasn't working until now. Also, I managed to fix some bugs which crop up with long sequences. Apologies to anyone who this affected.

    Finally, I've added an "Advanced Options" box to the bottom of the submission page. I've placed the Global-local option there along with a facility to turn off the low-complexity filter which by default marks with "X"s, regions of your sequence deemed to be low-complexity or coiled coil by the program "seg". If your results show your query sequence to contain "X"s, and you are certain the region in question is not low complexity, you can switch off the low-complexity filter and resubmit your sequence.
    Global Local Alignment Feb 04 2000
    I've placed an experimental new option on the fold recognition page. By default submissions will be treated as always - using a Global End gap penalisation dynamic programming algorithm. However, you may now choose Global-Local as an alternative technique. If your sequence is over 300 or 400 residues long I would recommend using this Global-Local option. Also, if a first attempt using the default does not make a confident assignment, it may be worth trying the Global-Local option.

    As I have said this is experimental, so any feedback would be greatly appreciated. Many people are using the server, and if users can tell me when they get a confident, sensible hit with one method as opposed to another, we can achieve some real-world benchmarking - and maybe learn something. Good Luck!
    E-mail Address Accuracy Jan 17 2000
    For some weeks now I have been receiving results for 3D-PSSM submissions that have bounced back to me because of problems with people's E-mail addresses.


    I have no way of getting these "lost" results to the submitter, as my only route to you is your E-mail address. If you do not receive any reply from the server when you have submitted your results, it is almost certainly because the E-mail address supplied is not valid. I would place these results in a "lost" section on the website, but I do not want to violate anyone's privacy of submission.

    If you do not receive your results within an hour and you know the descriptor you used, feel free to ask me (L.Kelley@icrf.icnet.uk) if I have the results.
    Downloadable Results Oct 15 1999
    It is now possible to download your html results in the form of a gzipped tar file. This contains all the models, multiple sequence alignments, etc. You can click either near the top of the html results page, or on your E-mail results. I know some people have wanted this functionality for some time. Its here at last.

    Also, I've fixed some submission problems. The server now tells you as soon as you click submit whether your sequence contains characters the server can't cope with. This is in contrast to having to wait 10 minutes only to receive an E-mail saying there's something wrong with the submission. I hope this is useful.
    Apologies for Broken Jobs      Sep 2, 1999
    Apologies to anyone who has had jobs killed since the re-opening of the server. Due to the extra functionality of the server, greater load is being placed on the machines. As a result, even relatively short sequences (300 residues) can sometimes cause our machines to grind to a halt. Work on fixing this is in progress. Please be patient, and in the meantime try to submit fragments of your sequence (100-200 residue chunks). Once again, apologies for the inconvenience. We will have this fixed a.s.a.p.
    New Server, New address, New functionality      Aug 25, 1999

    What's New?


  • Enlarged database (almost 3000 entries, previous size 2000).
  • Improved structural profiles.
  • Secondary Structure prediction by state-of-the-art PSI-Pred by David Jones.
  • Solvent Potentials to improve coverage and confidence.
  • Consensus of 3 scoring matrices.

  • Interface

  • Hopefully clearer, cleaner interface (more help files to come)

  • Power/Hardware

  • Fully fledged queuing system.
  • 4 processor cluster, moving to 6 cpus in a month or so.
  • Weekly updated sequence database

  • Future enhancements

  • Daily updated structure database
  • Pair-potentials

  • Please Note: I've just this moment released this to the community. It could fall over in any number of glorious ways. Use it at your own risk. I'm sure there are still some problems hiding in there, and I'll try to get everything smooth again a.s.a.p.. On another note, I have removed the facility for submitting PhD predictions. Its too much hassle trying to keep up with changes made to PhD output, and the secondary structure prediction here is now far better. It is currently NOT possible to submit a multiple sequence alignment in fasta format, but this will soon(ish) be remedied.

    Finally, I've just noticed some problems with our caching facilities. It looks as though the server is not working at top speed at the moment...a job for tomorrow.
    best wishes,

    Old News      Aug 25, 1999

    Previous News items for 3D-PSSM Version 1 can be found here