3D Variability Analysis (3DVA)

3DVA applies principal component analysis (PCA) directly to the 3D density maps of a particle stack. This decomposes structural variance into orthogonal variability components ordered by the amount of motion they explain. The first component captures the largest amplitude motion, while subsequent components resolve increasingly subtle dynamics. Unlike neural network-based approaches, the linear decomposition guarantees interpretability and avoids overfitting to noise.

Rather than sorting particles into discrete classes, 3DVA models a continuous conformational landscape. Each particle is assigned a coordinate along each variability component. By interpolating along a component's axis, users can generate a morph—a smooth, physically plausible movie of the molecule transitioning between extreme states. This reveals the full trajectory of domain rotations, hinge motions, and subunit rearrangements without artificial binning.

Traditional 3D classification struggles to resolve low-population states because they lack sufficient particle counts for independent reconstruction. 3DVA's PCA framework models the global covariance of the entire dataset, allowing it to detect rare conformations that represent less than 1% of particles. This is critical for capturing transient intermediates in enzymatic cycles or fleeting ligand-bound states.

3DVA offers a Simple Mode that performs variability analysis on a downsampled particle stack using a reduced number of components. This provides a rapid, low-resolution preview of the dominant motions in minutes, enabling iterative experimental feedback during data collection. Users can identify whether a sample exhibits significant flexibility before committing to high-resolution refinement.

While 3DVA is fundamentally a continuous method, its Cluster Mode applies k-means clustering to the latent coordinates to partition particles into discrete groups. Each cluster can then be independently reconstructed to high resolution. This hybrid approach combines the sensitivity of continuous analysis with the resolution benefits of traditional 3D classification, useful when distinct biochemical states coexist.

3DVA operates natively within the cryoSPARC ecosystem, accepting inputs directly from homogeneous or non-uniform refinements. The output variability components can be visualized interactively, and the resulting particle subsets or interpolated maps can be passed to downstream jobs like local refinement or atomic model building. This seamless integration eliminates the need for external scripting or format conversion.