Improving protein conformational sampling by using guiding projections

Abstract

Sampling-based motion planning algorithms from the field of robotics have been very successful in exploring the conformational space of proteins. However, studying the flexibility of large proteins with hundreds or thousands of Degrees of Freedom (DoFs) remains a big challenge. Large proteins are also highly-constrained systems, which makes them more challenging for standard robotic approaches. So-called "expansive" motion planning algorithms were specifically developed to address highly-dimensional and highly-constrained problems. Many such planners employ a low-dimensional projection to estimate exploration coverage and direct their search based on this information. We believe that such a projection plays an essential role in the success of these planners. This paper shows how the low-dimensional projection used by expansive planners can be tailored with respect to a given molecular system to enhance the process of conformational sampling. We introduce a methodology to generate an expert projection using any available information about a given protein. We evaluate this methodology on several conformational search problems involving proteins with hundreds of DoFs. Our experiments demonstrate that incorporating expert knowledge into the projection can significantly benefit the exploration process.