Fig. 4. CryoSieve prioritizes the removal of radiation-damaged particles.

a Particles were selected from micrograph movie stacks of the proteasome, with each stack containing 32 frames and a total electron dose of 50 e⁻Å⁻². This electron dose was uniformly distributed across all frames. Particles were extracted from consistent positions, using averages from frames 5–14, 10–19, 15–24, and 20–29. The average electron doses absorbed are denoted at the bottom, and four representative particles are displayed for each radiation damage level. b The graph depicts the proportions of particles with varying levels of radiation damage (differentiated by colors) in the retained particles across various retention ratios (indicated on the left). A comparison was made between the particles retained by cisTEM (left horizontal bars), CryoSieve (middle horizontal bars), and NCC (right horizontal bars). CryoSieve consistently sieved out particles in a sequence from high to low radiation damage, demonstrating superior performance over both cisTEM and NCC. c Particle distribution across the four radiation damage levels was analyzed using iteration 6 (featuring a retention ratio of 26.2%) from CryoSieve, NCC, and cisTEM. The analysis also incorporated particles retained by the AGC and non-alignment classification methods, with retention ratios auto-determined for these methods. d, e The side chains of the density maps reconstructed by CryoSPARC, using retained particles, were compared alongside the corresponding model-to-map FSCs. This comparison utilized a retention ratio of 26.2% (from iteration 6) for CryoSieve, NCC, cisTEM, and random methods. The retention ratios for AGC and non-alignment classification were auto-determined. The intersection between the FSC threshold (FSC = 0.5) and the FSC curve is represented as a vertical dashed line.