Protein folding by a biased Monte Carlo procedure in the dihedral angle space

Abstract

A straightforward method for predicting the protein structure is to find conformations that have the lowest energy along a chosen folding pathway. One approach in this direction is to produce a large number of structures by varying the dihedral angles of the molecule more or less randomly and then to screen each one using a suitable energy function. This procedure is computationally demanding, but by using a more realistic model, one hopes that the folding behavior one observes in calculations may better mimic the actual folding process in nature. The method is beginning to yield interesting results, thanks to the increase in the computational power but also to the intelligent selection of the folding pathway. This article reviews general features of this method, some important highlights of the particular procedure we used, and some of the more significant results obtained to date in our laboratory. The results are highly encouraging and indicate the direction of future effort that is most likely to be fruitful.

Ab initio protein structure prediction using physicochemical potentials and a simplified off-lattice model

“…Reducing the degrees of freedom in a protein model restricts conformational space, allowing a greater percentage to be searched in a given time. Current models range in complexity and include fully flexible backbone torsions folded by Monte Carlo 2 and by molecular dynamics, 3 restricted backbone geometries, 4 -6 refined geometries weighted by statistical probabilities, 7 and lattice framework models. 8 -11 Although all these models lead to inaccuracies in geometry that in turn will compromise the effectiveness of any energy function, studies by Park and Levitt 12 and Rooman et al 13 have shown that protein backbones may be modeled with reasonable accuracy by using restricted sets of four or six Ramachandranangle geometries.…”

“…Successful discrimination of the native fold has been achieved by using purely hydrophobic and polar pairwise potentials, 20 whereas other studies have found that a backbone hydrogen bond function is also required. 7 More complex models may use a broad mixture of local and non-local interaction parameters, 21 combinations of hydrophobic and hydrogenbonding functions with harmonic bond potentials, 22 protein-database potentials of mean force optimized by folding ability, 2 and force fields in which all atoms are represented explicitly. 6 As structural complexity increases, the force fields become less applicable to the structure prediction of larger proteins, and it is still not clear how detailed a representation is necessary.…”

A Descriptive Analysis of Populations of Three-Dimensional Structures Calculated from Primary Sequences of Proteins by Osiris

“…2-6 are secondary structures assigned by STRIDE: 2 is the Xray structure, 3 is the barycenter including in its population the Xray, and 4, 5, 6 are the gray barycenters. structures indicate optimal performances of the methods in their current version and good structures were published (Wilson and Doniach, 1989;Skolnick and Kolinski, 1990;Sun, 1993;Dandekar and Argos, 1994;Kolinski and Skolnick, 1994;Brasseur, 1995;Srinivasan and Rose, 1995;Sun etal, 1995;Gunn, 1996;Kurochkina and Lee, 1994;Lee et al, 1996;Rabow and Scheraga, 1996;Yue and Dill, 1996). However, to our knowledge, none of the previously reported procedures gives fully satisfactory results no matter what the sequence is.…”

“…Among theoretical approaches are direct calculations of 3D-structures of globular proteins from primary sequences and dihedral lattices. Various groups using these approaches have proposed computing procedures: OSIRIS, used in this work (Brasseur, 1993;Brasseur, 1995), GEOCORE (Sun, 1993;Sun etal, 1995;Yue and Dill, 1996); LINUS (Srinivasan and Rose, 1995) and other procedures (Gunn, 1996;Lee et al, 1996;Rabow and Scheraga, 1996). We shall focus our study on those approaches.…”

De Novo Protein Structure Prediction

“…Rather than using the torsional angles of real fragments, move sets are based on continuous distributions of angles derived from known structures. Unlike earlier attempts (Lee et al, 1996), we have developed a system to mimic three-residue fragment replacement taking into account secondary structure. For each possible three-residue sequence with each possible secondary structure, a continuous basis / angle distribution for the central residue is determined based on the observed angles in known structures.…”