All algorithms
3D classification
Stage
3
of 8
Included
3D classification
Sorts a particle stack into 3D classes to separate conformations, compositions and junk. It uses the same Bayesian framework and job settings as RELION Class3D, computed on our own GPU engine. Combined with our platforms abililty to scale to dozens of GPUs, our 3D classification provides extremely high throughputs that support the sorting of 100s of millions of particles per day for ML applications.
3.9x
faster compared to v2 & optimized Relion
20.5 M
particles sorted in 6.5h using 4 GPUs
What it does
After 2D classification, a particle stack still mixes conformational states, partial complexes and particles that will never align well. 3D classification splits the stack into a set number of classes, each with its own map, so you keep the classes you want and carry them into refinement. It is one of the longest jobs in a workflow, and it is often run several times per dataset with different class counts or masks.
CryoCloud Class3D cuts that time. We reran 17 RELION Class3D jobs on 14 datasets, from 62,000 to 5.1 million particles, with the same particles, reference and settings. The median job finished 3.9x faster. See the full benchmark, including all 81 class maps.
How it works
Each iteration compares every particle image with projections of every class map over a grid of orientations and shifts, assigns probabilities, and rebuilds the maps from the weighted images. CryoCloud Class3D keeps each class’s full 3D grid in GPU memory and runs the search, the reconstruction and classification.
Global and local search, fast subsets and Blush regularisation are supported. With Automated Class3D Convergence, the job stops as soon as the classes settle.

Resulting class average maps from a 3D classification job.
Where it sits
3D classification can be done directly after particle extration, or following 2D classification. Provided decoy maps or automatically generated ones from our Decoy Generator job can be added as classes to pull junk out, which is our recommended way for automated workflows. AutoClass3D then selects the classes to keep, and the selected particles go on to 3D Refinement.
At a glance
Stage
3D classification
Runs
After extraction or 2D classification
Hardware
GPU
Inputs
Particles
Outputs
Class maps
per-class resolution and particle share, particles with class assignments
Replaces
Class3D (RELION)
Availability
Included
If you want to quickly sort mllions of particles, you can try out CryoCloud for free via our 30 day trial.
Put this algorithm to work
See the full system around the workflow step, then choose the right way to get started.
TECHNOLOGY OVERVIEW
Explore all algorithms
Browse the complete analysis workflow, from pre-processing to refinement.
View algorithm overview
PLAN WITH CONFIDENCE
See what’s included
Compare plans, compute, and storage with transparent costs for every workflow.
Explore pricing
SEE YOUR DATA FLOW
Book a guided demo
Talk through your data, workflows, and deployment needs with a CryoCloud specialist.
Book a demo
