Ab initio reconstruction of highly complex cryo-EM mixtures

Resolve complex cryo-electron microscopy mixtures containing tens or more distinct molecular species at the ab initio reconstruction stage, beyond the capabilities of existing methods.

Background

The paper describes ab initio heterogeneous reconstruction as a particularly difficult problem because particle poses, class assignments, and multiple three-dimensional volumes must be inferred jointly without initial poses or reference volumes. Existing multiclass methods generally require a small, predetermined number of classes, while more expressive neural-field and mixture-model approaches remain difficult to optimize as mixture complexity increases.

The authors identify reconstruction of mixtures containing tens or more distinct species as unresolved and note that existing methods had not recovered the underlying structures from the Tomotwin-100 benchmark. The paper’s reconstruction-ensemble meta-algorithm is presented as a response to this challenge and demonstrates substantial progress on 45- and 100-class synthetic mixtures.

References

Despite this progress, resolving complex mixtures containing tens or more distinct species at the ab initio stage remains an open challenge, with no existing methods, to our knowledge, able to recover any of the underlying structures from the Tomotwin-100 benchmark dataset.

— A meta-algorithm for ab initio reconstruction of complex mixtures in cryo-EM  (2608.25388 - Kaz et al., 26 Aug 2026) in Section 2.3, “Heterogeneity analysis beyond discrete classes”