Figure 14.

Results for alanine tetrapeptide in vacuum at 300 K. FESs for the MRSE embeddings trained on the unbiased and biased simulation data. (a) PT embedding trained and evaluated on the PT simulation data. (b,c) WT-MetaD embeddings trained and evaluated on the WT-MetaD simulation data. The WT-MetaD embeddings are obtained without (b) and with (c) incorporating weights into the training via a reweighted feature pairwise probability distribution (see eq 9). The FESs are obtained using KDE as described in Section 3.4. The state labels in the FESs correspond to the labeling used in Figure 13a. The embeddings are rescaled so that the equilibrium states are of similar size. The units for the MRSE embeddings are arbitrary and thus not shown.