Distance-based Graph Linearization and Sampled Max-sum Algorithm for Efficient 3D Potential Decoding of Macromolecules

Three-dimensional structure prediction of a molecule can be modeled as a minimum energy search problem in a potential landscape. Popular ab initio structure prediction approaches based on this formalization are the Monte Carlo methods represented by the Metropolis method. However, their prediction performance degrades for larger molecules such as proteins since the search space is exponential to the number of atoms. In order to search the exponential space more efficiently, we propose a new method modeling the potential landscape as a factor graph. The key ideas are slicing the factor graph based on the maximum distance of bonded atoms to convert it to a linear structured graph, and the utilization of the max-sum search algorithm combined with samplings. It is referred to as Slice Chain Max-Sum and it has an advantage that the search is efficient because the graph is linear. Experiments are performed using polypeptides having 50 to 300 amino acid residues. It has been shown that the proposed method is computationally more efficient than the Metropolis method for large molecules.Three-dimensional structure prediction of a molecule can be modeled as a minimum energy search problem in a potential landscape. Popular ab initio structure prediction approaches based on this formalization are the Monte Carlo methods represented by the Metropolis method. However, their prediction performance degrades for larger molecules such as proteins since the search space is exponential to the number of atoms. In order to search the exponential space more efficiently, we propose a new method modeling the potential landscape as a factor graph. The key ideas are slicing the factor graph based on the maximum distance of bonded atoms to convert it to a linear structured graph, and the utilization of the max-sum search algorithm combined with samplings. It is referred to as Slice Chain Max-Sum and it has an advantage that the search is efficient because the graph is linear. Experiments are performed using polypeptides having 50 to 300 amino acid residues. It has been shown that the proposed method is computationally more efficient than the Metropolis method for large molecules.