A conceptually simple model for protein-folding phenomena has been created: it is two-dimensional and has only two different ‘‘amino acids.’’ Ground-state conformations have been determined for all of its flexible polypeptides containing seven or fewer monomers. This complete database displays a wide geometric variety of folded shapes and shows that single point mutations in some cases induce substantial folding modifications. Neural-network concepts have been employed to analyze results. The simplest static neural networks required to act as error-free read-only memories provide a way to visualize the logical structure of underlying folding principles. The topologies of optimal networks found thus far suggest that protein folding has a more complex cooperative character than has been embodied previously in theoretical approaches.

• Received 1 March 1993

DOI:https://doi.org/10.1103/PhysRevE.48.1469