| Popis: |
Side chain optimization is the process of packing the sidechains of a protein onto a fixed backbone structure, such that the energy of the resultant structure is minimized. The continuous space of sidechain conformations is typically handled by discretizing (sampling) into a finite set of representative conformations called a "conformer library". The key contribution of this work is to use machine learning methods to distribute (conformational) sampling among different positions in a protein. The idea is that different positions in a protein backbone have different sampling requirements, for example, solvent exposed positions require less sampling than positions in the core of a protein. We propose a 3-ary categorization of every position in a target protein based on its sampling requirements and evaluate it by comparing against an unbiased distribution of conformers. Our results demonstrate that this strategy helps to distribute the sampling more efficiently for sidechain optimization. |
