DGM ignores sometimes the latent variable
know parts of the proteins' states, and the corr relationships, how would other proteins's state probably be?