On-the-Fly Rotamer Pair Energy Evaluation in Protein Design

Andrew Leaver-Fay, Jack Snoeyink, Brian Kuhlman

Abstract: Most existing algorithms for protein design, including those in the Rosetta molecular modeling program, precompute energies for rotamer pairs, since these energies can be examined repeatedly. Simulated annealing algorithms, however, do not examine these energies with the same frequency; while some are examined many times, others may not be examined at all. This paper compares strategies for computing these energies on the fly and caching computed energy values that are likely to be reused. By avoiding the expense of computing pair energies that are not examined by simulated annealing, we show that some caching strategies not only improve running time in design, but also use 90% less memory, which allows design computations to be performed on memory-limited machines.