Sampling properties of the alternating projection distance geometry algorithm applied to unconstrained polypeptide chains

A comparison of the sampling properties of the alternating projection algorithm (APA) with those of other distance geometry algorithms for poly-L-alanine conformations is presented. The effects of several additions and modifications to the algorithm and the input data are studied. These include a parameter that controls the compactness of conformations and metrization, which chooses initial distances that satisfy the triangle inequalities. The effect of adding short-range distance bounds implied by steric interactions is also investigated. We find that with metrization our algorithm has similar sampling properties to the DISGEO program, as measured by mean square end-to-end distance, RMS coordinate deviation, and dihedral angle deviation. As an alternative to metrization, the compactness parameter can be used to sample conformations in a desired compactness range. We also find that improving short-range bounds leads to improved local geometry in the computed conformations with no additional computational cost. In appendices we show that the rms coordinate deviation is a metric on the set of equivalence classes of coordinate matrices varying by a rotation, and we present our pseudo code to compute random column metrization in O(n(3)) flops. Copyright (C) 1996 Elsevier Science Ltd