Despite the explosion in protein structural and functional data, our understanding of protein movement is still very limited.  Experimental methods cannot operate at the time scales necessary to record protein folding and motions, and traditional simulation techniques such as molecular dynamics and Monte Carlo methods are too computationally expensive to simulate long enough time periods for most proteins of interest.  However, it is critical that we better understand protein motion and the folding process for several reasons: understanding the folding process can give insight into how to develop better structure prediction algorithms, diseases such as Alzheimer's and Mad Cow disease are related to protein misfolding, and many biochemical processes are regulated by protein motion.

This project developed a novel technique for computing molecular motions that is based on the successful probabilistic roadmap (PRM) method for robotics motion planning.  We were inspired to apply these sampling-based motion planning methods to protein folding based on our prior success in applying them to paper folding.  These methods represent a trade-off between approaches such as molecular dynamics and Monte Carlo simulations that provide detailed individual folding trajectories and techniques such as statistical mechanical methods that provide global energy landscape statistics.  This research made several important algorithmic advances to the foundations of sampling-based motion planning and also in the development of parallel algorithms capable of exploiting massively parallel computers.  In particular, we developed novel algorithmic approaches that have advanced the state of the art in motion planning, including: (i) a family of new methods that solved a long standing open problem in the field by developing strategies for uniformly sampling surfaces embedded in high dimensional spaces; (ii) a new strategy for applying machine learning in heterogeneous environments that removes the need for explicitly partitioning the planning space as had been done in previous methods; (iii) a suite a decomposition based methods that enable efficient and scalable parallelization of sampling based motion planning methods for the first time.  These algorithms were applied to traditional robotics planning problems and to problems in computational biology, including mapping protein folding landscapes and to study ligand binding affinities, that is an important step in pharmaceutical drug design, and were shown to provide significant improvements over previous methods.

The publicly available folding server maintained as part of this project includes a database of molecular motions (http://parasol.tamu.edu/foldingserver/) that we have computed with our method.  We also process requests that other researchers submit to our server.  As a part of this project, the folding server was updated to support submissions that include multiple input conformations of the protein, e.g., known intermediate conformations or known initial and final states.

Training and the development of human resources was an important aspect of this project.  A number of graduate students, the majority from groups underrepresented in computing, were involved in this project. Two women (one Hispanic) have already received their PhD degrees, and two more (one Black) will be graduating in the coming year.  All of them are planning research careers, with three targeting academia and one a research lab.  A Black male also received a PhD and he is pursuing a research career in a research lab.  One Native American male received a masters degree and is now a PhD student.  Additionally, several undergraduates (including women and underrepresented minority males) and high school students (including a black male) have been trained on this proje...

Please report errors in award information by writing to: awardsearch@nsf.gov.