Questions about spisonet.py reconstruct with --indenpendent

[Wei Wang]

First of all, thank you very much for developing the software Spisonet. I tried using it to optimize preferred orientation and achieved significant results. Figure 1 shows the map before optimization, and Figure 2 shows the map after optimization using spisonet.py reconstruct. It is clear that the stretching problem has been significantly mitigated.

However, when I added the --independent parameter, the optimized map became very strange, as shown in Figure 3 and 4 with different views. The entire map turned lamellar and was severely stretched in one direction. To upload the optimized map to the EMDB, I need to be able to calculate FSC curves from half-maps.

Could you please help me understand what might be causing this issue and if there are any solutions? Additionally, I have also tried using misalignment correction with RELION to correct the preferred orientation, but it did not yield significant results. The most effective result I have obtained so far is by using spisonet.py reconstruct without the --independent parameter, but this result does not allow for FSC curve calculation, which is quite frustrating for me.

[YUNTAO LIU]

Hi Wei,

I currently do not know whether there is a bug or need different parameters for --independent. A systematic test for --independent is needed. 

The bottom line is that running spIsoNet with --independent will not yield a better map compared to not using this parameter. 

The current suggestion is that you can do the regular postprocessing with spIsoNet generated maps, using those to make figures or submit to EMDB as additional maps. For deposit half maps and FSC to EMDB, You may want to use the original maps and original FSC without spIsoNet treatment. 

 

--
You received this message because you are subscribed to the Google Groups "spIsoNet" group.
To unsubscribe from this group and stop receiving emails from it, send an email to spisonet+u...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/spisonet/d11eecde-0aa2-4c71-a1b5-9b61b5e6ffd9n%40googlegroups.com.

--

Best Regards,

Yuntao Liu,  Postdoc.

California NanoSystem Institute

University of California Los Angeles

[Wei Wang]

Hi yuntao,

Thank you very much for your suggestions. Actually this issue has troubled me for a long time. I have tried many different parameters and used different half-maps to run  spIsoNet with --independent, but the results were always similar.

Finally, I still have a small question. If I upload the half-maps generated by spIsoNet as additional maps, can I use these maps to build an atomic model? How reliable are the maps generated by spIsoNet?

[YUNTAO LIU]

Hi Wei,

Using the spIsoNet predicted maps for model building should be fine. Based on my experience, the metrics of model geometry should improve when using the spIsoNet anisotropy-corrected map. This reliability stems from the self-supervised nature of spIsoNet, which ensures that the additional information it predicts comes from your dataset.

Nevertheless, the --independent is actually a valuable parameter to evaluate how trustworthy the spIsoNet generated maps are, because that provides two fully independent measurements from the process. I will check the code to see whether there is any potential problem with --independent argument. 

To view this discussion on the web, visit https://groups.google.com/d/msgid/spisonet/843686b9-f400-4f82-bdcb-3c741a4f0fdfn%40googlegroups.com.

[Wei Wang]

Hi yuntao,

Thank you very much for your suggestions, and for developing the software spIsoNet. It has been of great help to my research! I also look forward to seeing further improvements in its functionality.