#35 · Structural bio & protein design

Anomaly detection and reconstruction-aware pipelines reshape cryo-EM particle picking

The dominant thread across these three preprints is handling contaminating and heterogeneous species during cryo-EM particle selection and reconstruction. CryoAnomaly trains an anomaly detector on clean synthetic cryo-EM data and uses it to suppress real contaminants as hard negatives in few-shot picking. A reconstruction-aware pipeline integrates picking, contamination removal, and 2D class selection, linking the picking step to downstream reconstruction. The meta-algorithm addresses complex mixtures at the reconstruction level, spawning and aggregating multi-class ab initio reconstructions and reporting accuracy on Tomotwin-100 subsets and recovery of ribosomal components. The new elements are anomaly-based negative suppression and the explicit coupling of picking with reconstruction rather than treating picking as an isolated preprocessing step. All results are preprint-level and not independently validated.