Neural Network and Secondary Structure Prediction

The model comprises three layers of processing units – the input layer, the output layer, and so-called hidden layer between these layers. Signals are sent from the input layer to the hidden layer and from the hidden alyer to the output layer through junctions between the units. This configuration is referred to as a feed-forward multilayer network.

<snip />

When hidden layers are included, a neural network model is capable of detecting higher levels of interactioon among amino acids that influence secondary structure. For example, particular combinations of amino acids may produce a particular type of secondary structure. To resolve these patterns, a sufficient number of hidden units is needed (Helley and Karplus 1991); the number varies from 2 to a range of 10-40. An interesting side effect of adding more hidden units is that the neural network memorizes the training set but at the same time is less accurate with test sequences.

David W.Mount, Bioinformatics, Sequence and Genome Analysis, Cold Spring Harbor Laboratory Press, pp.452-454, 2001

Similar Posts:

Post a Comment

Your email is never published nor shared.