Fig. 5: Comparison of CryoTRANS and CryoDRGN pseudo-trajectories using a pre-catalytic spliceosome dataset.

a The UMAP visualization of latent encodings (middle) after training a 10-dimensional latent variable with CryoDRGN on the spliceosome dataset (EMPIAR-10180). Five positions are pinpointed along a linear path (the trajectory of CryoDRGN) within the latent space, each representing a density map encoded by the VAE's decoder within CryoDRGN. The initial and target maps are defined as the density maps at the starting and ending points of CryoDRGN's trajectory, respectively. b The trajectories of CryoDRGN and CryoTRANS are juxtaposed for comparison, considering both front and side perspectives. For a clearer visualization of the movements, the contour of the density map is also displayed. A local magnification of the side view is showcased through a highlighted frame.