A limiting-case study of protein structure prediction: Energy-based searches of reduced conformational space

The accurate prediction of the 3-dimensional structure of a protein from it s sequence is a major unsolved problem. The possibility of such a predictio n using an atom-based energy function and a systematic search procedure has been demonstrated here in a model problem. Three proteins (parvalbumin, a fibronectin domain, and CheY) in different classes (all alpha, all beta, an d mixed alpha/beta, respectively) were studied. All the secondary structura l elements and the side chains were fixed in the X-ray conformation, and on ly one backbone dihedral angle in each of the loops between the secondary s tructural elements was allowed to vary. Energy-based searches of this reduc ed conformational space were carried out with a solvent-modified empirical energy function. At each stage of the searches, many structures were genera ted by varying different combinations of angles, and the minimum-energy str ucture was selected as the starting structure for the next stage. The final minimum-energy structures for all three proteins were within 0.24 Angstrom of the X-ray structures. Less extensive search protocols, in which only on e angle was varied at a time, became trapped. The energy surfaces were foun d to become steeper as the absolute energy minimum was approached. The resu lts support the inference from other studies that the energy function used here has its minimum at the native structure. In addition, they demonstrate the importance of using a search algorithm in which many structures, inclu ding variations in multiple degrees of freedom, are evaluated at each stage . Although the model calculations treat a highly simplified version of the protein folding problem, the methodology used here and its success provide insights that should aid in the study of more realistic models.