Parallel constrained minimization for interactive protein modeling

A coarse-grain parallel algorithm for nonlinear constrained minimization allows interactive modeling of electrostatic interactions in proteins. The constrained minimization problem is the basis of a graphical protein modeling system called Sculpt. While a user interactively changes a protein structure or explores protein-drug interaction, Sculpt maintains protein properties such as bond lengths and non-bonded atom separation. Between screen updates, Sculpt finds a focal energy minimum that satisfies constraints specifying ideal bond lengths and angles. This paper shows a parallel algorithm that models electrostatic interactions in 3500-atom proteins at two updates per second on an eight-processor SGI Challenge, a 7.2-times improvement over the sequential algorithm. These results are demonstrated through applications of Sculpt in biochemistry and drug design.<>