Protein Structure and Function: Application of Bioinformatics Methods - John Rigden 2014

Bioinformatics methods for studying the structure and function of disordered proteins
Prediction of disorder
Comparison of disorder prediction methods

As mentioned earlier, prediction Methods differ in performance and prediction quality, which was confirmed in the Critical Assessment of Structure Prediction (CASP) experiments, specifically CASP6 (Jin and Dunbrack 2005) and CASP7 (Bordoli et al. 2007). Naturally, the performance of Disorder Prediction methods depends on the evaluation criteria and the datasets used. In essence, two main factors limit direct comparison between methods: 1) comparison cannot be based solely on simple percentage values of disorder, because the number of true positives can easily be increased at the expense of false positives, i.e., predicting disorder in ordered regions; 2) The amount of available data for ordered and disordered structures differs significantly, complicating a straightforward comparison of prediction accuracy. Accordingly, the performance of prediction methods is compared in various ways, primarily by determining sensitivity and Specificity, i.e., The ratio of correctly identified disordered regions relative to misidentified ordered regions. To be fair, the aforementioned prediction methods perform comparably to the best Secondary structure prediction algorithms. To obtain a reliable assessment of disorder, it is recommended to use several Methods based on different principles.