Figure 6

Methodology for ab initio prediction. We start with a sequence and generate conformations using two different move sets: fragments from a database with identical sequence and a 14-state φ/ψ model. Many trajectories are generated and minimised using two different protocols: Monte Carlo with simulated annealing and a genetic algorithm search. The minimisation function is primarily an all-atom conditional probability discriminatory function, a hydrophobic compactness function, and a bad contacts function. Once a set of conformations is generated, a hierarchical filtering technique is applied using many different filters/scoring functions to produce one or a few final conformations.