Algorithm about 3D variability

I am curious about the new function in cryosparc 2.9: 3D variability.

From the tutorial I understand a 3D covariance matrix is constructed before performing PCA. Is the 3D covariance matrix being estimated using the algorithm in “Liao et al., 2015, Structure 23, 1129–1137” or or it is based on other references?

Warmest regards,

1 Like

Hi @wwangat, this algorithm is new and not published yet - it is relatively different from previous approaches. We do not construct the 3D covariance matrix nor use PCA directly. Instead only the top-K eigenvectors of the covariance are computed in an iterative fashion. We decided to release the feature in cryoSPARC immediately so that users could start taking advantage of it, but we are still working on characterizing and fully describing the method for publication.

Thank you so much for your kind reply, looking forward to the publication of the methods!

Hi Ali,

I tried 3D variability and it could capture some of the conformational changes of my protein. But we are wondering how it works. Have you published the paper regarding the 3D variability? Thank you very much.

Hi @bryangxl,
Unfortunately we still haven’t submitted the publication but 3D variability in cryoSPARC. Understanding the results and interpretation, however, is detailed here:

In general, you can think of the results as similar to several methods that have previously existed which attempt to compute 3D PCA or 3D Eigen-volumes of the dataset. The major difference in cryoSPARC is the algorithm under the hood that actually is used to solve the optimization problem in these settings.

Hope this helps for now!