Efficient generation of low-energy folded states of a model protein. II. Automated histogram filtering

Abstract

A number of short, Monte Carlo simulated annealing runs are performed on a highly frustrated 69-mer off-lattice model protein, consisting of a chain of 69 beads that are either hydrophobic, hydrophilic, or neutral in nature, and which demonstrably folds into a six-stranded β-barrel structure. We employ an iterative, consensus-based scheme to cluster the 725 nonbonded distances between the hydrophobic beads using, in tandem, Ward’s method for hierarchical clustering and k-means partitional clustering. We also independently analyze the same data using computer-automated histogram filtering, a technology designed to cluster high-dimensional data, without the tedium and subjectivity required by our iterative implementation of the two classical clustering methods. The memberships of low-energy clusters obtained from both classical clustering and automated histogram filtering approaches are remarkably similar. Nonbonded distance constraints are derived from these clusters and from small sets of the original unclustered conformations obtained by simulated annealing. Employing a distance geometry approach, we efficiently generate novel, low-energy conformations from each set of distance constraints, including the apparent native structure, up to 40 times faster than by doing additional simulated annealing runs. Over 33 000 unique locally optimized conformations are generated in total, substantially augmenting the number of low-energy states located by the original simulated annealing runs.