Part of Advances in Neural Information Processing Systems 3 (NIPS 1990)
Henrik Fredholm, Henrik Bohr, Jakob Bohr, Søren Brunak, Rodney Cotterill, Benny Lautrup, Steffen Petersen
Three-dimensional (3D) structures of protein backbones have been pre(cid:173) dicted using neural networks. A feed forward neural network was trained on a class of functionally, but not structurally, homologous proteins, us(cid:173) ing backpropagation learning. The network generated tertiary structure information in the form of binary distance constraints for the Co atoms in the protein backbone. The binary distance between two Co atoms was o if the distance between them was less than a certain threshold distance, and 1 otherwise. The distance constraints predicted by the trained neu(cid:173) ral network were utilized to generate a folded conformation of the protein backbone, using a steepest descent minimization approach.