Fig. 2: Demonstration of TS-DAR on the Müller potential for learning latent hyperspherical representations. | Nature Communications

Fig. 2: Demonstration of TS-DAR on the Müller potential for learning latent hyperspherical representations.

From: Exploring transition states of protein conformational changes via out-of-distribution detection in the hyperspherical latent space

Fig. 2

a The 2D-Müller potential. b Projections of molecular dynamics (MD) conformations onto the latent hyperspherical space, and the free energy (\(-{\mbox{k}}T{\mathrm{ln}}P(\theta )\)) is displayed, where \({\mbox{k}}T\) is the Boltzmann constant and temperature, \(P\left(\theta \right)\) corresponds to the probabilities of MD conformations at the polar angle \(\theta\) on the hypersphere. The dashed lines indicate the mean vectors of each of the three metastable states. c Visualization of the output state assignments (state 1 to 3) of the TS-DAR model overlaid on the Müller potential. d The same as (b) except that the state assignments rather than potential of mean force in the latent space are displayed. e and f are the same as (c) and (d), respectively, except that the results of a control experiment that only includes the VAMP-2 loss37 when training the TS-DAR model are shown. Source data are provided as a Source Data file.

Back to article page