Parallel Tree Search-based Protein Conformation Analysis System
Implemented on a Massively Parallel Computer
Makoto Ando, Yutaka Akiyama, Kentaro Onizuka and Tamotsu Noguchi
In this paper, we report on the design and the implementation of the
Parallel Tree Search-based Protein Conformation Analysis System. In
our protein conformation analysis model, we assume that the atoms in a
protein are hard spheres with certain radii, and the single covalent
bonds are turnable axes, each with a fixed length. Also, we assume
that the bond-angles are unchangeable. The conformational isomers are
generated by varying the torsion angles of all the turnable axes, and
each generated conformational isomer is checked for validity. We
employed an exhaustive search method for conformational searches, and
this gives our system an ability to analyze conformations even if the
whole physicochemical environment of the target protein segment is
unknown. Constraints on the range of distances between two specific
atoms, and constraints on the directions of bonds greatly reduce the
size of the search space. Using these constraints, our system
succeeded in analyzing the conformations of an opioid peptide, one of
the important peptides in drug design. The parallel version of our
program showed good speedups over the sequential version (192-fold
speedup using 256 PUs). We are convinced that our system can reduce
the cost and the length of time for the development of new drugs.