Particle sorting/cleaning after refinement

I was wondering what is the best way to proceed with 3D re-classification of particles in order to improve the overall 3D structure. Due to slight variability (also visible with the 3D variability tool), I want to identify sub-groups of particles/structures and proceed with a more homogeneous 3D refinement.

I tried for example 3D homogeneous refinement and 3D NU refinement and the imported those outputs in the heterogenous refinement, but since the structures were fairly similar, there was no improvement when I re-refined the corresponding structures and particle sets.

The tutorial, although detailed, does not cover cases where similar, but distinct, populations of particles exist in the sample, and how to proceed with those.

thanks for all the help,

Hi @stavros,

When you ran heterogeneous refinement, were you able to separate multiple conformations along the variability that you see?

In general, if there are small conformational changes present in a dataset, it’s likely that refining all the particles together will yield the highest average resolution, and especially in the regions of the molecule that do not change too much between conformations. After separating conformations, you should see improved structure quality in the regions that did have heterogeneity, but the overall resolution may not improve (since the number of particles available often drops significantly after splitting into multiple classes).

In your case, if heterogeneous refinement was unable to find multiple conformations, this could be because the conformational landscape is continuous, and so it is not well-explained my discrete classes (as in heterogeneous refinement) but is better explained by a continuous deformation (as in 3D Variability). In the latter case, local refinement (using a mask and optionally subtraction) is the currently best method for improving the refinement result, but will only work if the continuous deformation you are seeing is well modelled as a rigid body motion of some sub-region of the overall molecule.

We will try to improve the tutorials to show how users can deal with heterogeneity.