Non-uniform refinement: Adaptive regularization improves single particle cryo-EM reconstruction

Ali Punjani; Haowei Zhang; David J. Fleet

The preceding results indicate that non-uniform refinement improves particle image alignments during iterative refinement, yielding improved map resolution and quality. Here, we show that it also improves B-factors [25] of 3D reconstructions, indicating a more efficient use of signal from the same particle image data. Larger magnitudes (steep slopes) indicate greater signal lost. Given identical image data, differences in B-factor between uniform and non-uniform refinement indicate relative amounts of signal lost due solely to refinement. Nonuniform refinement generally has a significantly lower magnitude, indicating less less signal lost to noise in reconstruction at higher frequencies. Right: Guinier plot (log amplitude per frequency squared [25] ) for the Nav1.7 channel dataset described in Sec. 6.4.