journal menu
![[Figure 1]](/d/issues/2025/08/00/ic5124/ic5124fig1mag.jpg)
|
| Figure 1 Schematic summary of our method for probabilistic ab initio 3D reconstruction. Particles are grouped into 2D classes prior to execution of the ab initio 3D workflow and signal-enhanced 2D class averages are calculated. Following random initialization of the volume, particles are sampled in a balanced fashion across the 2D classes. In phase 1, greedy balancing is performed, selecting the particles that best agree with their corresponding class average, as measured by the noise-normalized Euclidean distance. In phase 2, the particles that rank among the 50% most similar to the class average are sampled randomly. In phase 3, the fraction subjected to sampling is increased to 85%. The matrix of probabilities is obtained through matching the sampled particles with re-projections of the volume in polar Fourier space, allowing fast low-pass limited rotational matching. In each iteration, one element of the matrix describes the probability that a sampled particle was obtained from re-projection of the reconstructed volume in a certain projection direction and in-plane rotation, obtained through discretization of the S2 (2-sphere) and S1 (circle) manifolds, respectively. |
![[Figure 1]](https://journals.iucr.org/10.1107/S2059798325005686/ic5124fig1.jpg)
 |
STRUCTURAL BIOLOGY |
ISSN: 2059-7983
access
