Guiding Search through Low-dimensional Projections

Robotics-inspired Search for Modeling Native Structure and Structural Transitions in Proteins

Highlight of work led by K. Molloyg, B. Olsong and A. Shehu* at BMC Struct Biol J 2013, IEEE/ACM Trans Comp Biol and Bioinf (TCBB) 2013, IEEE BIBM-W Comput Struct Biol Workshop (CSBW) 2012, ACM Conf on Bioinf and Comp Biol (BCB) 2012, J Bioinf and Comp Biol 2012, IEEE BIBM-W Comp Struct Biol Workshop (CSBW) 2011, J Bioinf and Comp Biol 2011, Intl. Conference on Bio-inspired Models of Network, Information, and Computing Systems (Bionetics) 2010, Intl J Robot Res 2010, Robot Sci and Sys Conf (RSS) 2009.

The protein conformational space is vast, and the underlying energy surface is rich in local minima. These two characteristics of proteins make it difficult for probabilistic search algorithms to have high exploration (sampling) capability. Specfically, it remains difficult to ensure that computed conformations are geometrically-distinct and not representative of few regions of the conformational space. The difficulty lies in the inability to find a few conformational (reaction) coordinates on which to define distance measures. Popular measures like least Root-Mean-Squared-Deviation (lRMSD) and radius of gyration (Rg), employed in our previous work on modeling native conformations of peptides and proteins, mask away differences and distort locations of regions deemed worthy of further exploration.

This new line of work pursues robotics-inspired search algorithms that capture the connectivity of the conformational space through global graph-like search structures. Lower-dimensional embeddings of the conformational space and underlying energy surface are employed to organize the search space and remember where the search algorithm has been. The centralized search structure allows efficiently computing global statistics and adaptively steering the search to low-energy under-populated regions in the search space, thus reapportioning limited computational resources where needed. Various realizations of this framework have been pursued to model native structures of proteins in the context of decoy sampling and also compute conformational pathways capturing motions of a protein system between two very different functional states.

On this Project:

  • Brian Olson

    Kevin Molloy

    Amarda Shehu

This material is based upon work supported by the National Science Foundation under Grant No. 1016995 and IIS CAREER Award No. 1144106. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.