DAMMIF, a program for rapid ab-initio shape determination in small-angle scattering

Abstract

DAMMIF, a revised implementation of the ab-initio shape-determination program DAMMIN for small-angle scattering data, is presented. The program was fully rewritten, and its algorithm was optimized for speed of execution and modified to avoid limitations due to the finite search volume. Symmetry and anisometry constraints can be imposed on the particle shape, similar to DAMMIN. In equivalent conditions, DAMMIF is 25-40 times faster than DAMMIN on a single CPU. The possibility to utilize multiple CPUs is added to DAMMIF. The application is available in binary form for major platforms.

Keywords: DAMMIF; DAMMIN; computer programs; particle shape determination; small-angle scattering.

Figure 1

Cross sections of dummy atom models of DAMMIN (left) and DAMMIF (right). The top row shows initial models (randomized and proto-particle) and the bottom row the final models by the two programs. The different colours indicate particle (red) and solvent (turquoise, blue and green) states of the dummy atoms. In DAMMIF, only red and turquoise beads are subject to phase changes; DAMMIN generally allows phase transitions anywhere in the search volume. Green solvent beads indicate the current, extensible, border of DAMMIF’s mapped area (all visible beads).

Figure 2

SA algorithm as implemented in DAMMIN (left) and DAMMIF (right). An initial starting model is refined to yield the best possible fit to the experimental data. In DAMMIN, only one neighbouring model is taken into account at a time. If multiple cores or CPUs are available, it is possible to prefetch multiple models in parallel, here shown as , , . Each prefetched model is then examined and either accepted or rejected, according to the rules of SA.

Figure 3

Reconstruction of a cylindrical particle with radius 10 Å and height 200 Å (bottom centre) from its simulated scattering pattern presented on the left-hand side [relative intensity I versus inverse ångströms; the distance distribution function p(r) computed by GNOM is displayed in the insert]. The starting (top row) and final (bottom row) models from DAMMIN and DAMMIF are displayed in the middle and right panels, respectively. DAMMIN ran in a slow mode inside the spherical search volume (packing radius Å, CPU time used 246 min). For DAMMIF, the value of was 3.0 Å and the run on the same single processor took 8 min.