Predicting Binding Regions within Disordered Proteins

• Disordered regions characteristics consistent with non-folding sequences: significant net charge, lack of aromatic residues, excess of hydrophilic groups
• Biases in PDB due to the existence of protein disorder

• This study uses two PONDRs: X-ray and calcineuran (CaN) neural network predictors (NNP), with a sliding window of 21

• X-ray NNP features: H C S W Y E D K Hydropathy Flexibility
• CaN NNP features: H C S W Y E V F R β-moment

• Cross prediction (predictions based on different NNP's training examples) has previously revealed different flavors of disorder (cite), making cross prediction useful for characterizing the similarities and differences of two predictors

• Cases presented show the NNPs are able to identify binding sites in disordered proteins by yielding false predictions of order

• Any given region could be in an equilibrium between order and disorder, having different values for the equilibrium constant ranging from predominantly disordered to predominantly ordered

• The differing compositional features of disordered proteins might provide a tool for protein function identification and help determine the function of identified disordered regions with no known function

• Functional regions in disordered domains might be recognized by local tendencies to form ordered structure