Fig. 4: Consensus landscapes obtained by FlexConsensus for the SARS-CoV-2 D614G spike variant.
From: Merging conformational landscapes in a single consensus space with FlexConsensus algorithm
Two conformational spaces were input to the FlexConsensus analysis, obtained from two independent runs of the following methods: HetSIREN in reconstruction mode5 and CryoDRGN2. Panel a compares the common consensus landscape (in grays) against each consensus subspace generated from each input method (shown in colors). The consensus landscapes show three main regions corresponding to three different conformational states of the spike RBD: three-down, one-up and two-up. Based on the consensus, it is possible to see that both methods correctly identify the three structural states. However, HetSIREN is more evenly distributed than CryoDRGN, which identifies two states more prominently. Panel b highlights the regions assigned a higher consensus error for each method. In both methods, the most unstable particles tend to organize in the periphery of the consensus conformational landscape, with the main regions of the landscape being more stable. SWD, sliced Wasserstein distance.
