The architecture and weights of an artificial neural network model tha
t predicts putative transmembrane sequences have been developed and op
timized by the algorithm of structure evolution. The resulting filter
is able to classify membrane/nonmembrane transition regions in sequenc
es of integral human membrane proteins with high accuracy. Similar res
ults have been obtained for both training and test set data, indicatin
g that the network has focused on general features of transmembrane se
quences rather than specializing on the training data. Seven physicoch
emical amino acid properties have been used for sequence encoding. The
predictions are compared to hydrophobicity plots.