
"Improved Prediction of Disulfide Connectivity from Protein Sequences"

Marc Vincent, Andrea Passerini, Matthieu Labbe', Paolo Frasconi

Machine Learning and Neural Networks Group
Dipartimento di Sistemi e Informatica
Universita' degli Studi di Firenze, Italy
email: {vincent, passerini, labbe, p-f}@dsi.unifi.it
Web: http://www.dsi.unifi.it/neural/

Details of the PDBselect data set:

- Folds: directory containing splitting information for the 10-fold cross validation.

- Sequences: directory containing the fasta files of the chains in the dataset.

- chainLabels: file containing the labelling information at the chain level. 
  One line per chain, displaying the following content:
  <chain label> <chain ID>

  <chain label> equals "1" if the chain contains at least one disulfide bridge, "-1" otherwise

- CysteinLabels: directory containing the labelling information at the cystein level. 
  One file per chain containing the list of the cystein labels. 
  The labels are ordered according to the position of the corresponding cysteins in the sequence.

  The cystein label equals "1" if the cystein is a half-cystine, "-1" otherwise