Classification of protein fold classes by knot theory and prediction of folds by neural networks: A combined theoretical and experimental approach

Abstract

We present different means of classifying protein structure. One is made rigorous by mathematical knot invariants that coincide reasonably well with ordinary graphical fold classification and another classification is by packing analysis. Furthermore when constructing our mathematical fold classifications, we utilize standard neural network methods for predicting protein fold classes from amino acid sequences. We also make an analysis of the redundancy of the structural classifications in relation to function and ligand binding. Finally we advocate the use of combining the measurement of the VA, VCD, Raman, ROA, EA and ECD spectra with the primary sequence as a way to improve both the accuracy and reliability of fold class prediction schemes.