Incollection,
How to combine amino acid sequences and neural networks to improve protein secondary structure prediction
Abstract Book of the XXIII IUPAP International Conference on Statistical Physics, Genova, Italy, (9-13 July 2007)
Abstract
One of the major goals in protein science is the prediction of the secondary structure of proteins starting from their amino acid sequence. This is a former step in solving the folding problem. We describe results based on the use of two neural networks, the former being a perceptron with one hidden layer, performing a supervised learning phase. Sequences of outputs from the first network are then introduced as inputs to the second network, which has a filtering effect. The networks are implemented in ANSI C language and are optimized to run on a cluster of workstations under the UNIX operating system with PVM protocol. Parallel implementation of the algorithms has been tested on the prediction of secondary and tertiary structure of proteins starting from their aminoacidic sequences, resulting in an overall accuracy better than 74% on a data base of 300 non redundant proteins. Moreover, a few important proteins, such as ribonuclease and mioglobin, are discussed in detail.
