Fig. 5

Effects of the particle-picking template used in FLC and the micrographs’ SNR on MLE optimization. Noisy micrographs showing influenza virus HA trimers with different SNRs were subjected to BOF testing, using different templates for particle picking. The corresponding SNRs of the micrographs from which the particle sets were picked were 0.005 (a, b and c), 0.002 (d, e and f), 0.001 (g, h and i) and 0.0005 (j, k and l). The templates used for particle picking were: a Gaussian circle (a, d, g and j), one projection view of the influenza virus HA trimer (b, e, h and k) and one projection view of the HIV-1 Env trimer (c, f, i and l). The particles picked by FLC were randomly divided into five classes and averaged. The resulting “class averages” are shown in the leftmost column of each panel (a-l). Each assembly of datasets was subjected to multi-reference MLE classification using the random class averages as starting references. In each panel, the five rows of image series correspond to five particle orientation classes generated by MLE, with the starting reference (S. Ref) and class averages of the milestone iterations (1st, 10th, 50th, and 100th) shown in a row. The BOF testing results show that MLE optimization can recover the weak signal of the influenza virus HA trimer if the images have a sufficiently high SNR