Extended Data Fig. 5: Analysis of those images estimated to have a larger consensus error in the SARS-CoV-2 dataset.
From: Merging conformational landscapes in a single consensus space with FlexConsensus algorithm
Reconstruction obtained from the SARS-CoV-2 dataset, considering only those images estimated to be within the 5% largest consensus error (around 21k particles from the original 440k). The landscapes show the distribution of these images in the consensus space: For CryoDRGN, most of these images are located in the two-up state region, while in HetSIREN, they are distributed along the regions corresponding to the one-up and three-down states. In contrast, the reconstruction calculated directly from these particles corresponds to a clear one-up state, indicating that the two heterogeneity methods wrongly assigned these images to regions of the conformational landscape.
