Predicting intrinsically disordered proteins based on multi-scale characteristics fusion

Due to the importance of functions, it has already become a hotter and hotter topic to predict intrinsically disordered regions in proteins. To consider the information from long and short disordered regions simultaneously and accurately predict both of the two regions, a new method based on multi-scale characteristics fusion was proposed in this article. First, characteristics based on different scales were extracted from amino acid sequences and used to build several basic models by SVM. Then the Q-statistics method was introduced to measure the diversity among all basic models. The basic models with the larger diversity were chose out and built the integrated predictor. Finally, majority voting method was used to make decision fusion and output the final predicting results. Subsequent simulation suggests that the proposed method can consider the information from the long and short disordered regions simultaneously and get a good predicting accuracy for IDPs, especially short disordered regions.