Predicting the disulfide bonding state of cysteines using protein descriptors

Knowledge of the disulfide bonding state of the cysteines of proteins is of major interest in designing numerous molecular biology experiments, or in predicting their three-dimensional structure. Previous methods using the information gained from aligned sets of sequences have reached up to 82% of success in predicting the oxidation state of cysteines. In the present study, we assess the relative efficiency of different descriptors in predicting the cysteine disulfide bonding states. Our results suggest that the information on the residues flanking the cysteines is less informative about the disulfide bonding state than about the amino acid content of the whole protein. Using a combination of logistic functions learned with subsets of proteins homogeneous in terms of their amino acid content, we propose a simple prediction approach, starting from a single sequence, that reaches success rates close to 84%. This score can be improved by avoiding predictions regarding cysteines for which the decision is not well marked. For example, we obtain a score close to 87% correct prediction when we exclude predicting 10% of the cysteines.