Cryo-EM AI Agent

Ariadne

An agentic thread through the Cryo-EM labyrinth: plans, executes, and refines reconstruction workflows end-to-end, delivering publication-ready structures from raw data.

Request Early Access See Capabilities

Capabilities

Drives the Workflow
Autonomously

Ariadne handles the full arc of a Cryo-EM experiment — not as a pipeline of scripts, but as an agent that understands your research goal and works backward from it.

01 — REASONING

Thinks like a scientist

Plans and reasons before execution — forming hypotheses, anticipating decisions, and designing failure strategies based on your specific research objective.

Hypothesis-driven

02 — ORCHESTRATION

Works across specialized tools

Intelligently orchestrates Cryo-EM software — selecting methods, adjusting parameters, and chaining tools as part of a coherent scientific strategy.

Multi-tool agent

03 — AUTONOMY

Set it in motion

Start the workflow once — Ariadne takes it from there, surfacing discoveries and advancing your experiment while you focus on higher-level science.

Fully autonomous

04 — OUTPUT

Publication-ready results

Closes the workflow into structured results — generating figures, summaries, and publication-ready materials directly from your processing history.

Research-grade

Demonstrations

Work With It

Give Ariadne a raw dataset and a target. It autonomously sequences the full reconstruction pipeline — from motion correction and CTF estimation through particle picking, 2D classification, and 3D refinement — without a single manual step.

Point Ariadne at an existing project and it reads the complete job graph — every preprocessing step, classification run, and refinement branch — then distils the processing logic into a named, reusable workflow template that can be stored and replayed across future datasets.

Point Ariadne at a reference project directory and it reads all stored workflow templates collectively, cross-references their processing strategies, and synthesises a new tailored reconstruction pipeline that draws on the full breadth of prior work.

Ariadne compiles the full processing history into a machine-readable structured report — particle counts, resolution metrics, and a visual workflow diagram — ready to share, archive, or hand off to the next stage of analysis.

Chat Interface
ariadne — terminal
Reconstruction Output Complete
Input Data 7,689 movies Motion + CTF 7,689 mics Import Particles 1,966,430 ptcls Extract · 336 px 1,873,811 ptcls 2D Class × 2 · Select 672,053 ptcls Ab Initio K=3 3 volumes Hetero Refine 3 classes
DIMER
J595 · NU-Refine · C1
3.75 Å
159,361 ptcls
FSC 0.143 · auto-tightened mask
MONO-A
J611 · NU-Refine · C1
4.55 Å
98,204 ptcls
FSC 0.143 · monomer conformation
MONO-B
J623 · NU-Refine · C1
4.43 Å
125,415 ptcls
FSC 0.143 · monomer conformation

From raw data to structure. No manual workflow.

Customized Workflow · P745 W17 Complete
J548 import_movies 3,845 movies J549 import_movies 3,844 movies J550 patch_motion 3,845 mics J551 patch_motion 3,842 mics J552 patch_CTF 3,845 exp J553 patch_CTF 3,842 exp J554 import_particles 1,966,430 ptcls J561 extract · box 336 px 1,873,811 ptcls J562 2D_class K=200 1,845,138 ptcls J563 select_2D 1,138,900 ptcls retained J564 2D_class K=100 curated J566 select_2D 672,053 ptcls J567 ab_initio K=3 initial volume generation J568 hetero_refine 3 classes class0 class1 class2 J574+J577 220K · ab_initio+refine J575+J578 268K · ab_initio+refine J576+J579 184K · ab_initio+refine J595 NU-Refine · C1 DIMER 3.75 Å J611 NU-Refine · C1 MONO-A 4.55 Å J623 NU-Refine · C1 MONO-B 4.43 Å
Reference Workflow · /projects/apo_form/ Ready
Reference Directory
▶ apo_form/
├── workflow_apo_v1.json WF
├── workflow_apo_v2.json WF
├── workflow_apo_v3.json WF
└── workflow_holo_v1.json WF
Workflow Synthesis
workflow_apo_v1 8 steps workflow_apo_v2 10 steps workflow_apo_v3 9 steps workflow_holo_v1 7 steps Collective Synthesis cross-referencing all strategies · synthesising synthesised pipeline 01 import_movies synthesised 02 patch_motion + patch_CTF synthesised 03 extract · 2D_class K=200 · select synthesised 04 ab_initio K=3 · hetero_refine synthesised NU-Refine · C1 Pipeline ready · 8 steps queued
Structured Output · P745 Generated
Output Files
▶ P745/
├── summarize.md MD
├── P745_presentation.pptx PPTX
└── P745_paper.pdf PDF
Processing Summary

A total of 7,689 movies (3,845 + 3,844 from two import batches) were processed with patch motion correction and patch CTF estimation in CryoSPARC. 1,966,430 particles were imported and re-extracted in a 336-pixel box (1,873,811 particles). Two rounds of 2D classification (K=200, then K=100) with manual selection reduced the dataset to 672,053 particles.

Ab initio reconstruction (K=3) followed by heterogeneous refinement separated particles into three classes: class 0 (220,387 particles, dimer), class 1 (268,027 particles), and class 2 (183,639 particles). Each class was re-extracted at 360-pixel box size.

Dimer (J595): Class 0 particles underwent ab initio (K=2) and heterogeneous refinement, selecting 159,361 particles. After re-extraction (box 320px), 2D classification (K=100), and selection (124,629 particles), final re-extraction at 360-pixel box yielded 124,504 particles. Non-uniform refinement (C1) achieved 3.75 Å (FSC 0.143, auto-tightened mask). 3D variability analysis was performed downstream.

Monomer branches: Class 1 particles (268,027) and class 0 rejects (59,145) were pooled (324,915 total), processed through ab initio (K=3) and heterogeneous refinement, selecting 164,849 particles. After re-extraction (box 256px), 2D classification, and selection (248,866 particles), two parallel paths were pursued:

  • Monomer-A (J611): Ab initio (K=2) → heterogeneous refinement → 2D classification (K=50) → selection (163,327) → ab initio (K=2) → heterogeneous refinement → non-uniform refinement of 98,204 particles at 4.55 Å (C1).
  • Monomer-B (J623): Ab initio (K=3) → heterogeneous refinement → non-uniform refinement of 125,415 particles at 4.43 Å (C1).

Early Access

Ready to let
Ariadne do the science?

We are onboarding a select group of academic labs and research groups. If you are pushing the frontier of structural biology, we would like to hear from you.

Request Early Access Work With Us