A Neural Network for Predicting Protein Disorder using Amino Acid Hydropathy Values

Proteins have been discovered to contain ordered regions and disordered regions, where ordered regions have a defined three-dimensional (3D) structure and disordered regions do not. While in the past it was believed that proteins only function in a defined 3D structure, proteins with disordered regions have been discovered to have at least 28 distinct functions. It is now important to be able to determine the ordered and disordered regions in proteins. Several experimental techniques such as X-ray crystallography, NMR spectroscopy, circular dichroism, protease digestion, and Stokes radius determination, along with several computational techniques such as artificial neural networks (ANNs), support vector machines (SVMs), logistic regression, and discriminant analysis have so far been used to detect disordered proteins. Past research has shown that ANNs and amino acid properties are an effective tool at predicting protein disorder. This research uses a feed-forward neural network implemented using JavaNNS and the hydropathy values of amino acids to predict protein disorder. The results show that hydropathy is an important amino acid property for disorder.