Protein fold class prediction using neural networks with tailored early-stopping

Predicting the three-dimensional structure of a protein from its amino acid sequence is an important problem in bioinformatics and a challenging task for machine learning algorithms. We describe an application of feed-forward neural networks to the classification of the protein fold class given the primary sequence of a protein. Different feature spaces for primary sequences are investigated, a tailored early-stopping heuristic for sparse data is introduced, and the achieved prediction results are compared to those of various other machine learning methods.