Sequence-Based Prediction of Molecular Recognition Features in Disordered Proteins

Abstract

Molecular recognition features (MoRFs) act as molecular switches in molecular-interaction network of the cell, and assumed to have relationship with the causes of many diseases. The importance of identifying MoRFs in disordered proteins is becoming increasingly apparent. So far, only a limited number of experimentally validated MoRFs is known, and there are few specialized tools for identifying MoRFs. Existing methods used many predicted results, such as predicted disorder probabilities, solvent accessibility and B-factors as features for prediction, or used MoRFs database directly for alignment to assist the prediction; however, their design are complex, and the performance is also affected largely by other predictors. In this study, we proposed a novel method, named as MFPSSMPred (Masked and Filtered PSSM based Prediction), which adopts a masking method to extract high local conservative features, and a filtering method to filter out low local conservative scores in position-specific scoring matrix (PSSMs) for prediction. All features are extracted from the sequences only. We compared our method with a traditional PSSM-based method and 9 other existed methods on a same test dataset. The experimental results showed that, our method achieved the best performance with AUC of 0.758. This study demonstrated that: 1) the flanking regions of MoRFs affected the plasticity of MoRFs; 2) MoRFs were flanked by less conserved residues; and 3) the revised PSSM was predictive features for identifying