A tree-based search to bias sampling of protein decoy conformations

In this study we investigate a probabilistic search framework for its ability to reduce redundancy in its sampling of the decoy ensemble and obtain a broad view of the energy surface. The framework is inspired by robotics motion-planning algorithms and naturally allows further analyzing the role of energy in biasing the sampling of low-energy decoy conformations. In addition to a local energy bias that through the Metropolis criterion leads the search over time to low-energy conformations, the framework implements a global energy bias. A search tree maintains all decoy conformations. The tree grows by selecting a conformation and expanding it to obtain a new one. A new conformation is the result of a short Monte Carlo trajectory that employs molecular fragment replacement. Energies conformations in the tree are projected on a id grid, which allows then implementing different selection techniques and tuning the strength of the global energy bias to steer sampling towards a particular distribution of decoys. The selection techniques make use of weighting functions defined over grid cells.