Question about the inital models for refinement

[dahui-y]

Hi!

In step 6, Reference-based refinement, we need to input an initial model for refinement. I am wondering how we can obtain the initial model if we process another dataset. I am trying to reproduce the result of EMPIAR-10499. I am wondering if the initial model is the result of subtomogram averaging, which we can also find on the EMPIAR website.

Thanks

---

17 Jan Updated
Dear Prof. Abartesaghi,

Extend this question: if we would like to investigate things with unknown structures in cellular, which cannot be found in the EMDB, how do we obtain the initial reference model?

Thanks

Originally posted by @dahui-y in #134

[abartesaghi]

Apologies for the delayed response, this message got overlooked since it wasn’t on the main discussion page.

If an external reference (from EMDB or elsewhere) is unavailable, there are two options for performing ab-initio reconstruction:

1. Standard sub-volume averaging
2. 2D projection-based ab-initio reconstruction

Both methods are available in nextPYP (starting with version 0.7.3, which is currently in beta, not officially released).

For sub-volume averaging, the approach involves iterative alignment and classification of sub-tomograms until classes with the expected features emerge. This process is interactive, similar to 2D classification in SPA, where you’ll need to manually select good classes from bad during each iteration and adjust alignment parameters depending on the shape and size of your target.

For 2D projection-based reconstruction, alignment and classification (optional) are done using raw 2D particle projections. While this is more automated, it is also more prone to errors and may not work depending on the complexity of your target.

In both cases, it is crucial to start with the cleanest particle set possible.