Skip to main content

Advertisement

Fig. 7 | BMC Bioinformatics

Fig. 7

From: Deep clustering of protein folding simulations

Fig. 7

CVAE learned features predict folding intermediates across two independent folding simulations of BBA protein. a depicts the CVAE latent space embedding on one 223 μs trajectory of BBA folding. Each conformation from the simulation is projected onto a three dimensional embedding and painted with the fraction of native contacts. One can observe that folded and unfolded states are separated into distinct clusters. b To elucidate the embedding from the CVAE, we project the conformations from the trajectory using t-SNE and identify conformational states captured by the CVAE. States are captured as described in Fig. 5 and shown as a cartoon representation based on the cluster from which they belong to. These conformational states depict various levels of BBA folding, labeled (i) through (vi). c Using the model learned on the longer trajectory, we project the conformers from the second independent simulation of length 102 μs onto the same latent space. It is notable that the folded states are similarly clustered together while the unfolded states are captured separately. d The latent representation from the CVAE can be applied across different trajectories to summarize folded states. As shown in the cartoon representation, the conformational ensembles are separated into distinct folded (labeled (ii)), partially folded (labeled (i) and (iii)) and fully unfolded states (labeled (iv) and (v))

Back to article page