Constrained template matching using rejection sampling
Identifying macromolecular complexes in situ using cryo-electron tomography remains challenging, with low signal-to-noise ratios, the missing wedge, and crowded backgrounds among the key limiting factors. By integrating prior knowledge on macromolecular localization, such as the preferred orientations of membrane-associated proteins, detection can be improved by constraining searches to biologically feasible orientations. Here we describe rejection sampling, an approach for integrating such constraints at voxel resolution that remains both accurate and computationally efficient. Using synthetic and experimental data, we show that these constraints improve detection, orientational assignment, and discrimination between macromolecules. The resulting picks match the performance of deep-learning methods informed by membrane structure without requiring annotated data. We further apply rejection sampling to ATP synthase on mitochondrial cristae, illustrating how it extends macromolecular detection to the large and highly curved membrane systems that pervade cells.