Ab-initio yield better map than NU-refinment on a small membrane protein complex

Dear all,
I am working on a small membrane protein complex ~130kDa. The complex is flexible and after sorting I end up with a small set of particles (70k). It yields a reconstruction at 6A resolution (NU- refinement new) with a map consistent with this resolution. Nonetheless, the weak quality of the trans-membrane regions in comparison with other regions of the density map reflects remaining flexibility. I subjected the particles to an ab-initio job with two models, to discriminate the best particles (limit of resolution 6A), like in this paper “Structure and drug resistance of the Plasmodium falciparum transporter PfCRT https://doi.org/10.1038/s41586-019-1795-x”.
One of the two Ab-initio models (accounting for 44k particles) displays a nice overall density and feature trans-membrane regions with better qualities than the ones from precedent NU-refinement new. I subjected this new dataset (44k particles) to a new round of NU-refinement new. Unfortunately, it yields a low-resolution 6,5A and the density looks really poor (around 10A resolution).
Do you guys have any suggestions of parameters I can play with during refinement or any other suggestions?
Thank you for your help,

  • For very small proteins, NU-refine can be sensitive to the initial lowpass, and the default can be a bit aggressive. Reducing to 15 or even 12Å may give better results, if the initial ab initio model is already pretty good

  • Try proceeding directly to local refinement (with NU regularization on) from the orientations in the ab initio job.

  • Try NU-refine with a static (and very soft) mask, rather than using the default dynamic masking parameters.


Dear Oli,
Many thanks for your help. NU-Refine with a reduced lowpass filter give significativity better results.
Best ,

You should also consider that ab initio is not using gold-standard refinement, therefore you basically double the available particle number. Especially for little datasets with low SNR particles that can cause a considerable difference.


Hi Olibclarke,
What do you mean by soft mask?

A mask with a softer edge, e.g 9 pixels

Thanks Olibclarke for your answer. Is the “initial lowpass” = lowpass filtered reference? L

yes - in cryosparc the reference is filtered on the fly before the beginning of the job. The degree to which it is filtered is controlled by this parameter

Thanks again! when we do the particle pick, should we decrease the lowpass filter for very small particles? (to 15 A or lower?)

No - I would be very careful about using templates with high resolution features for picking. You risk ending up with Einstein-from-noise classes, introducing bias from the template into your data.

1 Like

thanks for your fast reply!

Hi Oli, would you explain the parameter “enforce non-negativity” in NUR? when to use?

Dear Tarek,
Thank you for your response. Sorry for the late response I missed your email. It’s a fair point, nonetheless, I suppose that all the data are merged together in the final reconstruction like in relion. But as you said it can be deleterious in the particles orientations determination if the SNR is too weak in precedent iterations.