OPUS-DSD

OPUS-DSD reconstructs dynamic structural landscapes from cryo-electron microscopy (cryo-EM) single-particle images to characterize conformational heterogeneity of macromolecules.

Key Features:

• Three-Dimensional Convolutional Encoder-Decoder Architecture: Employs a three-dimensional convolutional encoder-decoder neural network trained on cryo-EM images to capture structural variations and encode high-dimensional data into a compact representation.
• Low-Dimensional Structural Landscape Mapping: Maps structural variations into a smooth, analyzable low-dimensional space that can be traversed continuously to reconstruct dynamic processes or identify discrete states.
• Clustering and Resolution Enhancement: Clusters similar particles within the learned representation to improve reconstruction resolution and separate distinct conformational states.

Scientific Applications:

• Enzyme conformational dynamics: Analysis of enzymes undergoing conformational changes during catalysis to link structural ensembles with function.
• Signaling protein conformational ensembles: Characterization of proteins involved in signaling pathways to resolve multiple functional states.
• Highly dynamic macromolecular complexes: Dissection of structural heterogeneity in complexes with intrinsic dynamism to identify continuous and discrete conformations.

Methodology:

Train a three-dimensional convolutional encoder-decoder neural network on cryo-EM single-particle datasets to encode high-dimensional volumes into a smooth low-dimensional latent space, then traverse and cluster that space to reconstruct continuous dynamics and discrete conformational states.