Octopi and TopCup particle picks for a self-configuring in situ cryoET structure-determination workflow

Deposition Overview

Particle picks produced by Octopi and TopCup, two deep-learning particle-detection models from the self-configuring in situ cryoET structure-determination workflow (Octopi + py2rely). Octopi is a self-configuring 3D U-Net whose architecture/hyperparameters are selected per dataset by Bayesian optimization; TopCup is an open re-implementation of the top-ranked 2024-2025 Kaggle CryoET Object Identification Challenge solution, used as a fixed-architecture baseline. Picks are deposited for: GroEL in in-situ E. coli; apo-ferritin, beta-amylase, beta-galactosidase, ribosome, thyroglobulin and virus-like particle in the Phantom/ML-Challenge datasets (both Octopi and TopCup); cytosolic and membrane-bound ribosomes in unroofed cells; and V-ATPase on isolated synaptic vesicles. All picks are model predictions validated by subtomogram averaging with py2rely (RELION-5).