Protein Fold Recognition Using Neural Networks

Lin, Guang (2003). Protein Fold Recognition Using Neural Networks. PhD thesis The Open University.

DOI: https://doi.org/10.21954/ou.ro.0000f713

Abstract

To predict accurately the three-dimensional (3D) structures of proteins from their amino acid sequences alone remains a challenging problem. However, using protein fold recognition tools, it is often possible to achieve good models or at least to gain some more information, to aid scientists in their research. This thesis describes development of TUNE (Threading Using Neural Networks), a fold recognition program using artificial neural network (ANN) models.
A new method to generate amino acid substitution matrices is described in chapter two. It uses an ANN to generalise amino acid substitutions observed in protein structure alignments. Matrices for alignment scoring from this approach were compared with classic alignment scoring schemes.
From these neural network models, a series of encoding schemes were constructed. These schemes describe the amino acid types with a few numbers. They were generated to replace the orthogonal encoding scheme, so that smaller, faster and more accurate neural network models can be applied on bioinformatic problems.
The TUNE model was introduced in chapter four to measure protein sequence-structure compatibility. Given the integrated residue structural environment descriptions, the model predicts probabilities of observing amino acid types in such environments. Using this model, a scoring function to measure the fitness of a residue in a protein structure model can be made for protein threading programs.
The model in chapter two was extended by including the residue structural environment descriptions for predictions. A simple protein fold recognition program with a dynamic programming algorithm was developed using this model. The program was then tested in the fourth round of the Critical Assessment of protein Structure Prediction methods (CASP4) and produced reasonably good results.