Skip to main content


Fig. 2 | BMC Bioinformatics

Fig. 2

From: Deep clustering of protein folding simulations

Fig. 2

Quantifying CVAE performance over MD simulations. We quantify the loss, \(\mathcal {L}\), described in the “Results” section over the number of training epochs for each of the systems, namely a Fs-peptide, b VHP, and c BBA. As the number of epochs increases, the loss function decreases, indicating convergence in the training process. We also quantify the quality of reconstruction based on the CVAE’s ability to predict contacts within the testing dataset. For each of the simulations, namely d Fs-peptide, e VHP, and f BBA, we show the reconstruction difference between the actual and predicted contact matrices from the testing data and histogram it to quantify how many contacts may be mispredicted by the model. See text for additional details

Back to article page