Automated detection of disulfide bridges in electron density maps using linear discriminant analysis

The ability to recognize disulfide bridges automatically in electron density maps would be useful to both protein crystallographers and automated model‐building programs. A computational method is described for recognizing disulfide bridges in uninterpreted maps based on linear discriminant analysis. For each localized spherical region in a map, a vector of rotation‐invariant numeric features is calculated that captures various aspects of the local pattern of density. These features values are then input into a linear equation, with coefficients computed to optimize discrimination of a set of training examples (disulfides and non‐disulfides), and compared with a decision threshold. The method is shown to be highly accurate at distinguishing disulfides from non‐disulfides in both the original training data and in real (experimental) electron density maps of other proteins.