Answers are grounded in this paper and cite their sources.
Cryo-electron microscopy (cryo-EM) datasets preserves a wide range of structural heterogeneities in macromolecules, arising from both large-scale dynamics and compositional changes. The deep learning framework OPUS-DSD has been developed to characterize the structural heterogeneities in cryo-EM data. However, different types of variations often entangle together in its results. To address this, here we propose OPUS-DSD2, an algorithm which incorporates dynamics model into OPUS-DSD to enhance the disentanglement of large-scale dynamics and compositional differences. OPUS-DSD2 retains the convolutional decoder from OPUS-DSD to represent compositional changes and fine-scale dynamics through 3D volumes. Additionally, it introduces a dynamics decoder based on multi-layer perceptron to model large-scale dynamics via deformation fields. OPUS-DSD2 employs separate latent spaces for training these decoders. Comparative tests on real datasets have shown that OPUS-DSD2 effectively separates dynamics and compositional changes into distinct latent spaces, and encodes compositional changes in different regions of macromolecules to different principal components of composition latent space. OPUS-DSD2 integrates with the original OPUS-DSD framework, and is available at https://github.com/alncat/opusDSD.
No takes yet. Share an insight, caveat, or question.
Luo et al. (2024) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: