Distinct Structural Flexibility within SARS-CoV-2 Spike Protein Reveals Potential Therapeutic Targets

Serena H. Chen; M. Todd Young; John Gounley; Christopher Stanley; Debsindhu Bhowmik

To gain further insight into the 20,000 protomer structures, we applied convolutional variational autoencoders (CVAEs) to encode the high dimensional protein structures from the MD simulations into 3-D latent spaces for visualization. Compared to commonly applied structural analysis methods, which are mostly limited to specific regions of the protein of interest, our DL approach allows us to evaluate a protein structure as a whole yet still capture detailed local structural features. We found individual clusters corresponding to each of the four protomer systems, suggesting that the structural features of the four systems embedded in the latent spaces are distinct from each other ( Fig. 2 A-C). Representative structures selected from the clusters also show high structural flexibility in their domain arrangement ( Fig. 2 D-F). The hinge region connecting the S1 and S2 subunits opens up, causing broad distributions of the interdomain distances observed in Fig. 1 E-G.