The first set of protocols (Table I ) we considered relax the sidechains but keep the backbone fixed. Sidechains are optimized in two steps—first, discrete combinatorial rotamer optimization and second, continuous optimization of the sidechain torsion angles. The combinatorial rotamer optimization (referred to as repacking throughout the remainder of the text) is carried out using Monte Carlo simulated annealing with the Dunbrack backbone dependent rotamer library.9 (link) The continuous optimization is carried out using quasi-Newton minimization and is referred to as minimization throughout the remainder of the text.
We experimented with two energy functions at both the repacking and minimization steps. The first is the standard Rosetta all atom energy function used in prediction and design calculations;10 (link) we refer to this as “hard-rep” because the Lennard-Jones repulsive interactions are not damped, thus atomic clashes incur very large energetic penalties. The second has the repulsive interactions at short atomic separations damped as described in theSupporting Information but is otherwise identical; we refer to this as “soft-rep” because small atomic overlaps are not heavily penalized.
We also experimented with allowing different numbers of residues surrounding the site of mutation to be repacked. As indicated in theTable I protocol summary, we considered three possibilities: first, only repacking the mutated residue, second, only residues within 8 Å of the mutated residue, and third, all residues.
We also explored protocols which carry out backbone torsion angle minimization following sidechain repacking in attempts to more accurately model the structural consequences of mutations. To prevent the backbone from moving too much from the native structure, in some protocols, we included distance constraints during the backbone minimization as described in theSupporting Information .
Finally, we explored protocols which more extensively search through alternative backbone conformations. We developed a Monte Carlo simulated annealing protocol that generates backbone conformations with ideal bond lengths and bond angles that uniformly sample the space of conformations surrounding any given native structure. The protocol carries out 100,000 moves each consisting of a small random perturbation of the backbone torsion angles; the scoring function prevents sampling from deviating by more than a specified tolerance from the starting structure. Single side chain rotamer flips are attempted at one-tenth the frequency of backbone moves. The resulting structures have small and partially compensating changes in nearly all the backbone torsion angles. The lowest energy structure sampled during each trajectory is subjected to backbone and sidechain minimization using the hard-rep energy function. Full details are provided in theSupplementary Information .
We experimented with two energy functions at both the repacking and minimization steps. The first is the standard Rosetta all atom energy function used in prediction and design calculations;10 (link) we refer to this as “hard-rep” because the Lennard-Jones repulsive interactions are not damped, thus atomic clashes incur very large energetic penalties. The second has the repulsive interactions at short atomic separations damped as described in the
We also experimented with allowing different numbers of residues surrounding the site of mutation to be repacked. As indicated in the
We also explored protocols which carry out backbone torsion angle minimization following sidechain repacking in attempts to more accurately model the structural consequences of mutations. To prevent the backbone from moving too much from the native structure, in some protocols, we included distance constraints during the backbone minimization as described in the
Finally, we explored protocols which more extensively search through alternative backbone conformations. We developed a Monte Carlo simulated annealing protocol that generates backbone conformations with ideal bond lengths and bond angles that uniformly sample the space of conformations surrounding any given native structure. The protocol carries out 100,000 moves each consisting of a small random perturbation of the backbone torsion angles; the scoring function prevents sampling from deviating by more than a specified tolerance from the starting structure. Single side chain rotamer flips are attempted at one-tenth the frequency of backbone moves. The resulting structures have small and partially compensating changes in nearly all the backbone torsion angles. The lowest energy structure sampled during each trajectory is subjected to backbone and sidechain minimization using the hard-rep energy function. Full details are provided in the
