# Job: Homogeneous Ab-Initio Refinement (BETA)

## At a Glance

Use CryoSPARC's *ab-initio* algorithm (based on stochastic gradient descent) to refine a dataset to high resolution from scratch, while preserving gold standard independence of two half-sets and half-maps.

## Description

{% hint style="warning" %}
Homogeneous Ab-Initio Refinement should generally be tried only once traditional global refinements fail to produce reasonable results. See Recommended Alternatives for more information.
{% endhint %}

In [some challenging cases](https://www.biorxiv.org/content/10.1101/2025.09.08.674935v1), refinement of a 3D map to high resolution using CryoSPARC's [gradient descent](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-reconstruction/job-ab-initio-reconstruction#stochastic-gradient-descent) algorithm for *ab-initio* reconstruction outperforms the traditional [expectation maximization algorithm](https://guide.cryosparc.com/expectation-maximization-in-cryo-em) used in refinement jobs \[1]. However, directly using the [Ab-Initio Reconstruction](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-reconstruction/job-ab-initio-reconstruction) job to produce high-resolution volumes makes resolution estimation and validation difficult, since Ab-Initio Reconstruction does not use gold standard half-sets.

**New in CryoSPARC v5.0,** This job splits input particles into two half-sets and refines independent half-maps each using the same process as [Ab-Initio Reconstruction](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-reconstruction/job-ab-initio-reconstruction). Notably, this job is still *ab-initio* in that it does not use any input volume; both half maps are reconstructed independently from scratch from the data alone, and kept aligned to each other by a 3D map alignment step within the job. This job can only produce a single homogeneous output rather than multiple classes.

The output half-maps can be further refined downstream using [Local Refinement](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/local-refinement/job-new-local-refinement-beta) which can sometimes further improve map quality. Note that using a global refinement downstream, like [Homogeneous](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-refinement/job-homogeneous-refinement) or [Non-Uniform Refinement](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-refinement/job-non-uniform-refinement-new), typically will not yield results superior to using standard Ab-initio Reconstruction followed by refinement.

## Inputs

### Particle stacks

Particles from which half maps will be produced.

## Commonly Adjusted Parameters

See [Ab-Initio Reconstruction](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-reconstruction/job-ab-initio-reconstruction#commonly-adjusted-parameters) for parameter information.

## Outputs

### Particles

Particles with poses and half-set splits used to produce the half maps.

### Volumes

Unlike other refinement jobs in CryoSPARC, Homogeneous Ab-Initio Refinement produces only half maps.

### Unused particles

Particles which were not used. This typically means that the refinement converged before all particles had been aligned. If you wish to align all particles, set `Num particles to use` to a value greater than or equal to the number of input particles.

## Common Next Steps

To produce an averaged, filtered full map you can either use [Homogeneous Reconstruction Only](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-refinement/job-homogeneous-reconstruction-only) with the particles output, or use the particles and half maps in [Local Refinement](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/local-refinement/job-new-local-refinement-beta) with `Initialize from input half-maps` turned on.

## Recommended Alternatives

Homogeneous Ab-Initio Refinement currently requires far more iterations than other techniques to produce a final map. For this reason, this job should only be used when both:

* Particle images produce high-quality 2D classes, indicating that real signal is present in the particle stack
* Global refinements of particles (e.g., [Homogeneous](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-refinement/job-homogeneous-refinement) or [Non-Uniform Refinement](https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-refinement/job-non-uniform-refinement-new)) produce maps with a quality much worse than expected based on 2D Classification results

## References

1. Kookjoo Kim, Huan Li, Oliver B. Clarke (2025). High-resolution ab initio reconstruction enables cryo-EM structure determination of small particles. bioRxiv 2025.09.08.674935; doi: <https://doi.org/10.1101/2025.09.08.674935>


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://guide.cryosparc.com/processing-data/all-job-types-in-cryosparc/3d-reconstruction/job-homogeneous-ab-initio-refinement-beta.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.