DISULFIND is a server for predicting the disulfide bonding state of cysteines and their disulfide connectivity starting from sequence alone. Disulfide bridges play a major role in the stabilization of the folding process for several proteins. Prediction of disulfide bridges from sequence alone is therefore useful for the study of structural and functional properties of specific proteins. In addition, knowledge about the disulfide bonding state of cysteines may help the experimental structure determination process and may be useful in other genomic annotation tasks.
Our server predicts disulfide patterns in two computational stages: (1) the disulfide bonding state of each cysteine is predicted by a BRNN-SVM binary classifier; (2) cysteines that are known to participate in the formation of bridges are paired by a Recursive Neural Network to obtain a connectivity pattern.

• Predict both the bonding state of each cystein in the query and the corresponding disulfide arrangment. Option: Predict bonding state + connectivity pattern.
• Predict only the disulfide pattern of some of the cysteines. Option: Predict connectivity from user specified bonding state
  Note: as you might imagine, this last option represents few possible situations:
  • You know the bonding state of each cysteine in the chain and would like to know how the oxidized cysteines are connected.
  • You know the bonding state of some of the cysteines and would like to know how they would be connected.
  In case the user selects this option, after pressing the submit button, a page will be displayed with a list of checkboxes, each one representing the position of a cysteine in the submitted query. Please: do not make use of the navigation/reload buttons at this stage.
  After pressing the submit button, the user selected checkboxes will trigger the prediction of connectivity among the corresponding cysteines only. Errors will be feedbacked in case of obvious inconsistent selections, e.g. odd or zero number of selected cysteines.

DISULFIND employs an SVM binary classifier to predict the bonding state of each cysteine [2,3], followed by a refinement stage that collectively classifies all the cysteines in the chain. Global dependencies represented by the collective bonding state are modelled in two ways. Firstly, the SVM receives as input both local and global descriptive attributes. Secondly, we add a post-processing stage consisting of a bidirectional recurrent neural network (BRNN) in the attempt of recovering some correlation, and a Viterbi decoder that finds the maximum likelihood bonding state assignment which satisfies the constraint of having an even number of disulfide bonded cysteines (interchain bonds are ignored).

Predicting disulfide patterns can be seen as mapping an input sequence with annotated cysteines into an output graphs representing disulfide connectivity. The input is formed by the annotated sequence and a candidate connectivity pattern. The method applied in DISULFIND, documented in [1], employs a bi-directional Recursive Neural Network trained to predict the expected distance of candidate patterns to the real one. Prediction is carried out by running the trained network on all possible connectivity patterns and choosing the one yielding maximum score.

Your email address, the place where the prediction will be delivered, if option Batch: send output to email address is selected.
NOTE: Check that you typed your address correctly.

An optional name for your query. We strongly suggest that you use one, especially if sending more than one query. The order in which you send your queries may not correspond to the order in which you receive the answers.

The sequence of aminoacids:

• A bare sequence is accepted. Please no FASTA format.
• Spaces, newlines and tabs will be simply ignored.
• Non alphabetical chars will cause the rejection of the query.
• Only 1 letter amino acid code accepted. Please do not send nucleotide sequences. If so, A will be treated as Alanine, C as Cysteine, etc...

Replies are sent either by email (option: Batch: send output to email address) or as an html page (option: Interactive: receive output on the browser). In the latter case, after pressing the send button, an intermediate page will appear and auto-refresh every 10 seconds until the final output is returned.
By default DISULFIND only returns the most likely connectivity pattern. By setting the number of alternatives to an integer k in the range [2,3], the k best ranking patterns will be returned.

EMAIL output example

Query Name: VPRA_DENPO


                          +----------------------+
                     +----|------------+         |                    +---------------+
               +-----|----|+           |         |                  +-|-----+         |
               |     |    ||           |         |                  | |     |         |
         .........10........20........30........40........50........60........70........
AA       AVITGACERDLQCGKGTCCAVSLWIKSVRVCTPVGTSGEDCHPASHKIPFSGQRKMHHTCPCAPNLACVQTSPKKFKCL
DB_state       1     1    11           1         1                  1 1     1         1
DB_conf        9     9    99           9         9                  9 9     9         9
DB_flip





         80
AA       SK
DB_state
DB_conf
DB_flip

DB_bond   bond(7,19)
DB_bond   bond(13,31)
DB_bond   bond(18,41)
DB_bond   bond(60,68)
DB_bond   bond(62,78)

Conn_conf 0.442914

----------------------------------------------------------------------------------------

SECOND best ranking connectivity pattern


                     +----+                      +------------------+ +---------------+
               +-----|----|+           +---------|------------------|-|-----+         |
               |     |    ||           |         |                  | |     |         |
         .........10........20........30........40........50........60........70........
AA       AVITGACERDLQCGKGTCCAVSLWIKSVRVCTPVGTSGEDCHPASHKIPFSGQRKMHHTCPCAPNLACVQTSPKKFKCL




         80
AA       SK

DB_bond   bond(7,19)
DB_bond   bond(13,18)
DB_bond   bond(31,68)
DB_bond   bond(41,60)
DB_bond   bond(62,78)

Conn_conf 0.422348

----------------------------------------------------------------------------------------

Abbreviations used:

AA        = amino acid sequence
DB_state  = predicted disulfide bonding state
            (1=disulfide bonded, 0=not disulfide bonded)
DB_conf   = confidence of disulfide bonding state prediction (0=low to 9=high)
DB_flip   = an asterisk (*) indicates that the viterbi aligner overruled the
            most likely predition for that residue in order to achieve a
            consistent prediction at a protein level (even number of disulfide
            bonded cysteines, as interchain bonds are ignored).
DB_bond   = position in sequence of a pair of cysteines predicted to be forming
            a disulfide bridge.
Conn_conf = confidence of connectivity assignment given the predicted disulfide
            bonding state (real value in [0,1])

----------------------------------------------------------------------------------------

HTML output example

                          +----------------------+                                      
                     +----|------------+         |                    +---------------+ 
               +-----|----|+           |         |                  +-|-----+         | 
               |     |    ||           |         |                  | |     |         | 
         .........10........20........30........40........50........60........70........
AA       AVITGACERDLQCGKGTCCAVSLWIKSVRVCTPVGTSGEDCHPASHKIPFSGQRKMHHTCPCAPNLACVQTSPKKFKCL
DB_state       1     1    11           1         1                  1 1     1         1 
DB_conf        9     9    99           9         9                  9 9     9         9 

           
           
           
           
         80

AA       SK
DB_state   
DB_conf    

DB_bond   bond(7,19)
DB_bond   bond(13,31)
DB_bond   bond(18,41)
DB_bond   bond(60,68)
DB_bond   bond(62,78)

Conn_conf 0.442914

                     +----+                      +------------------+ +---------------+ 
               +-----|----|+           +---------|------------------|-|-----+         | 
               |     |    ||           |         |                  | |     |         | 
         .........10........20........30........40........50........60........70........
AA       AVITGACERDLQCGKGTCCAVSLWIKSVRVCTPVGTSGEDCHPASHKIPFSGQRKMHHTCPCAPNLACVQTSPKKFKCL

           
           
           
         80
AA       SK

DB_bond   bond(7,19)
DB_bond   bond(13,18)
DB_bond   bond(31,68)
DB_bond   bond(41,60)
DB_bond   bond(62,78)

Conn_conf 0.422348

Please cite:

[1] A. Ceroni, A. Passerini, A. Vullo and P. Frasconi. DISULFIND: a Disulfide Bonding State and Cysteine Connectivity Prediction Server, Nucleic Acids Research, 34(Web Server issue):W177--W181, 2006. Download PDF

For the disulfide connectivity predictor see also:

[2] A. Vullo and P. Frasconi. Disulfide connectivity prediction using recursive neural networks and evolutionary information, Bioinformatics, 20, 653-659, 2004. Download PDF

For the cystein bonding state predictor see also:

[3] P. Frasconi, A. Passerini, and A. Vullo. A two-stage SVM architecture for predicting the disulfide bonding state of cysteines, Proc. IEEE Workshop on Neural Networks for Signal Processing, pp.25-34, 2002. Download PDF

[4] A.Ceroni, P.Frasconi, A.Passerini and A.Vullo. Predicting the disulfide bonding state of cysteines with combinations of kernel machines, Journal of VLSI Signal Processing, 35, 287-295, 2003. Download PDF

Copyright notice