Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Cryoem AI Drug Design Agent · FreedomIntelligence | DecimalAI

Cryoem AI Drug Design Agent

SkillScore77Maintained

cryoem-ai-drug-design-agent

AI-powered integration of cryo-EM structural data with generative AI and molecular dynamics for structure-based drug design targeting flexible proteins and membrane complexes.

scienceby FreedomIntelligence

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Research pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/cryoem-ai-drug-design-agent/SKILL.md

decimalai skills pull cryoem-ai-drug-design-agent --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/cryoem-ai-drug-design-agent/SKILL.md https://app.decimal.ai/s/cryoem-ai-drug-design-agent/SKILL.md

Raw file: /s/cryoem-ai-drug-design-agent/SKILL.md ↗ · machine-readable summary at /s/cryoem-ai-drug-design-agent.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 50d ago

Yes — +23% lift · at higher cost+42% tokens0% turns77% of whole cases pass with skill (17/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-03	✗→✓	▲ Improved	7%	0%
case-06	✗→✓	▲ Improved	50%	0%
case-10	✗→✓	▲ Improved	56%	0%
case-11	✗→✓	▲ Improved	62%	0%
case-16	✗→✓	▲ Improved	13%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/1/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

FreedomIntelligence / OpenClaw-Medical-Skills

skills/cryoem-ai-drug-design-agent/SKILL.md

View on GitHub ↗

Imported 8/1/2026Version v2

Skill contents

SKILL.md

<!--

COPYRIGHT NOTICE

This file is part of the "Universal Biomedical Skills" project.

Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu>

All Rights Reserved.

This code is proprietary and confidential.

Unauthorized copying of this file, via any medium is strictly prohibited.

Provenance: Authenticated by MD BABU MIA

-->

Cryo-EM AI Drug Design Agent

The Cryo-EM AI Drug Design Agent integrates cryo-electron microscopy structural data with AlphaFold3, generative AI, and molecular dynamics for structure-based drug design. It enables targeting of previously "undruggable" proteins including flexible, membrane-bound, and large macromolecular complexes through high-resolution structure-guided optimization.

When to Use This Skill

When designing drugs against cryo-EM-solved targets.
For fragment-based drug discovery with EM structures.
To model ligand binding in flexible protein regions.
When targeting membrane proteins and large complexes.
For integrating AlphaFold predictions with experimental EM density.

Core Capabilities

Density-Guided Design: Fit ligands into cryo-EM density maps.

AlphaFold Integration: Combine AF3 predictions with EM data.

Flexible Docking: Account for protein dynamics in binding.

Fragment Screening: Virtual fragment screening with EM structures.

Complex Targeting: Design for multi-protein assemblies.

Dynamics-Based Design: Incorporate conformational flexibility.

Cryo-EM for Drug Discovery

| Target Class | Cryo-EM Advantage | Drug Discovery Application | |--------------|-------------------|---------------------------| | GPCRs | Native lipid environment | Allosteric sites | | Ion Channels | Multiple conformations | State-specific design | | Transporters | Conformational states | Mechanism-based | | Ribosomes | Antibiotic binding | New antibiotics | | Viral Proteins | Large assemblies | Vaccines, antivirals | | Intrinsically Disordered | Flexible regions | Challenging targets |

Workflow

Input: Cryo-EM density map, protein sequence, ligand/fragment.

Structure Refinement: AlphaFold + density-guided refinement.

Binding Site Identification: Detect pockets in EM structure.

Ligand Placement: Density-guided ligand fitting.

MD Simulation: Flexible binding simulation.

Optimization: Generative design around hits.

Output: Optimized ligands, binding models, design recommendations.

Example Usage

User: "Design ligands for this GPCR cryo-EM structure, accounting for receptor flexibility in the binding pocket."

Agent Action:

bash
python3 Skills/Structural_Biology/CryoEM_AI_Drug_Design_Agent/design_from_cryoem.py \
    --density_map gpcr_3.2A.mrc \
    --protein_sequence gpcr.fasta \
    --alphafold_model gpcr_af2.pdb \
    --resolution 3.2 \
    --ligand_screening fragment_library.sdf \
    --binding_site_residues "3.32,5.46,6.48,7.39" \
    --md_refinement true \
    --generative_optimization true \
    --output gpcr_drug_design/

Input Requirements

| Input | Format | Purpose | |-------|--------|---------| | Density Map | MRC/MAP | EM density | | Protein Sequence | FASTA | AlphaFold input | | Resolution | Float (Å) | Quality metric | | Ligand Library | SDF | Virtual screening | | Known Ligand | Optional SDF | Starting point |

Output Components

| Output | Description | Format | |--------|-------------|--------| | Refined Structure | EM + AF combined | .pdb | | Ligand Poses | Density-fitted poses | .sdf | | Binding Scores | Affinity predictions | .csv | | Optimized Compounds | Generative designs | .sdf | | MD Trajectory | Flexibility analysis | .xtc | | Design Report | Recommendations | .pdf |

AI/ML Components

Structure Prediction:

AlphaFold3 for initial model
Density-guided refinement
Confidence scoring (pLDDT, local resolution)

Ligand Design:

Generative AI (diffusion, VAE)
Reinforcement learning optimization
Multi-objective scoring

Dynamics Integration:

Molecular dynamics simulation
Ensemble docking
Flexibility-aware scoring

Resolution Considerations

| Resolution | Applications | Limitations | |------------|--------------|-------------| | <3.0 Å | Fragment screening, detailed design | Rare | | 3.0-4.0 Å | Drug optimization, binding mode | Most targets | | 4.0-5.0 Å | Pocket identification, scaffold | Less detail | | >5.0 Å | Architecture, general binding | Low for SBDD |

AlphaFold3 + Cryo-EM Integration

| Scenario | Approach | Benefit | |----------|----------|---------| | Missing Loops | AF3 prediction | Complete structure | | Flexible Regions | Ensemble models | Multiple conformations | | Low Resolution | AF3 template | Higher confidence | | Ligand Binding | AF3 complex prediction | Binding mode |

Prerequisites

Python 3.10+
AlphaFold3, ChimeraX
GROMACS/OpenMM for MD
RDKit, AutoDock Vina
GPU with 16GB+ VRAM

Related Skills

Time_Resolved_CryoEM_Agent - Dynamics from EM
PROTAC_Design_Agent - Degrader design
Molecular_Glue_Discovery_Agent - Glue design
AlphaFold3_Agent - Structure prediction

Fragment-Based Discovery with Cryo-EM

| Step | Method | Cryo-EM Role | |------|--------|--------------| | Fragment Screening | Virtual dock to EM | Density-guided | | Hit Identification | Cryo-EM soaking | Experimental validation | | Fragment Growing | EM + modeling | Structure guidance | | Lead Optimization | Iterative EM | Binding mode confirmation |

Membrane Protein Targets

| Target Type | Cryo-EM Advantage | Examples | |-------------|-------------------|----------| | GPCRs | Native membrane | Numerous drugs | | Ion Channels | State-dependent | Painkillers, antiepileptics | | Transporters | Mechanism insight | Cancer, infection | | Receptors | Complex structures | Immunotherapy |

Special Considerations

Resolution Limits: Design confidence depends on resolution
Map Quality: Local resolution varies across structure
Conformational States: Multiple states may be captured
Ligand Density: May be weak at lower resolution
Validation: Experimental validation essential

Quality Metrics

| Metric | Purpose | Threshold | |--------|---------|-----------| | Global Resolution | Overall quality | <4.0 Å for SBDD | | Local Resolution | Binding site quality | <3.5 Å preferred | | Map Correlation | Model-to-map fit | >0.8 | | Real-Space R | Atomic fit | <0.3 | | Ligand CCC | Ligand fit | >0.6 |

Drug Discovery Success Stories

| Drug | Target | Cryo-EM Role | |------|--------|--------------| | Numerous | GPCRs | Structure determination | | Antibiotics | Ribosome | Binding mode | | Antivirals | Spike protein | Epitope mapping | | Various | Ion channels | State-specific design |

Author

AI Group - Biomedical AI Platform

Illustrative Want to see what cryoem-ai-drug-design-agent would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/2/2026, 17:56 UTC · 22 cases

Lift

+23%

77% vs 55%

Agent turns

1.0 vs 1.0

Tokens

+42%

3,855.09 vs 2,713.77

fail → pass · 5pass → pass · 12fail → fail · 5

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I have a 3.2 Å resolution cryo-EM density map (`gpcr_3.2A.mrc`), sequence FASTA (`gpcr.fasta`), and an initial AlphaFold prediction (`gpcr_af2.pdb`). I want to screen our fragment library (`fragment_library.sdf`) against active pocket positions 3.32, 5.46, 6.48, and 7.39. Please handle density-guided fitting, molecular dynamics relaxation, and generative optimization. Output the final refined protein-ligand PDB structure, fitted pose SDF file, CSV table of binding affinities, generative candidate SDFs, MD trajectory XTC file, and a summary PDF design report.	pass→pass	30,284	18,403	-39%	1	1	0%	6,291	5,658	-10%	0	0	—
case-02 We recently obtained a 3.8 Å cryo-EM map (`channel_em.map`) of a viral ion channel and its protein sequence (`channel.fasta`). We want to dock a small molecule library (`lead_library.sdf`) into the channel pore while incorporating AlphaFold3 structural modeling to reconstruct unresolved flexible loop regions. Please carry out density-constrained docking, perform flexible conformational sampling, and generate derivative analogs. We require the output files to include the merged PDB model, docked pose files, numerical binding scores in CSV format, optimized analog molecules, simulation trajectories, and a full design recommendation document.	fail→fail	24,130	27,148	+13%	1	1	0%	4,576	6,982	+53%	0	0	—
case-03 I am working on fragment-based drug discovery targeting a membrane transporter complex. I have the reconstructed electron density file `transporter_map.mrc` at 3.0 Å, sequence file `transporter.fasta`, and a target fragment collection `fragments_set.sdf`. Could you integrate AlphaFold3 prediction with the experimental map density to identify promising binding pockets, run MD-based flexible docking, and produce optimized lead compounds? Please deliver the refined PDB complex, candidate poses in SDF, CSV scoring spreadsheet, generated compound structures, trajectory files, and a comprehensive evaluation report.	fail→pass	26,665	19,306	-28%	1	1	0%	5,200	5,556	+7%	0	0	—
case-04 When evaluating whether a fitted atomic model is sufficiently accurate against a 3.2 Å electron density map before running generative compound optimization, our team usually aims for map correlation. What is the standard minimum threshold for map correlation in density-guided structure refinement?	pass→pass	12,306	5,504	-55%	1	1	0%	2,168	2,702	+25%	0	0	—
case-05 We fitted a ligand into a local density blob in a 3.5 Å receptor map. We need to decide if the ligand's density fit is strong enough to proceed with hit-to-lead expansion. What is the recommended Ligand Cross-Correlation Coefficient (CCC) threshold to confirm a valid ligand fit?	pass→pass	14,651	8,963	-39%	1	1	0%	2,442	3,159	+29%	0	0	—
case-06 During atomic model refinement into a reconstructed map, we monitor the Real-Space R factor. What maximum Real-Space R score threshold indicates acceptable atomic fit for structure-based drug design?	fail→pass	8,981	2,900	-68%	1	1	0%	1,522	2,278	+50%	0	0	—
case-07 A project team wants to initiate structure-based drug design on a membrane receptor structure solved at 4.2 Å global resolution. Is this global resolution adequate for high-confidence structure-based drug design, or what is the maximum recommended global resolution threshold?	pass→pass	13,086	7,531	-42%	1	1	0%	2,322	3,116	+34%	0	0	—
case-08 Although the overall density map of our target complex is 3.8 Å, the local resolution around the binding pocket varies. What local resolution threshold is preferred in the binding site to support confident ligand placement?	pass→pass	12,610	7,389	-41%	1	1	0%	2,078	3,070	+48%	0	0	—
case-09 We are setting up the automated pipeline script for a cryo-EM ligand fitting run with molecular dynamics relaxation. Which file extension should be used to record the molecular dynamics simulation trajectory?	pass→pass	6,711	2,858	-57%	1	1	0%	1,168	2,180	+87%	0	0	—
case-10 Our cryo-EM facility delivered a low-resolution map at 5.5 Å for a large macromolecular assembly. The computational team wants to perform high-resolution fragment docking and atomic sidechain optimization directly into this map. How should this map resolution (>5.0 Å) be handled regarding SBDD capabilities?	fail→pass	17,686	14,484	-18%	1	1	0%	2,645	4,123	+56%	0	0	—
case-11 We have an intermediate cryo-EM map at 4.5 Å resolution. What primary drug discovery tasks are appropriate at this 4.0–5.0 Å resolution range, rather than detailed atomic ligand optimization?	fail→pass	13,968	10,901	-22%	1	1	0%	2,170	3,515	+62%	0	0	—
case-12 Our cryo-EM map of a kinase complex at 3.1 Å resolution has several unassigned regions corresponding to flexible activation loops that lack clear electron density. How should predictive structural modeling be integrated to complete the target protein model?	pass→pass	17,041	14,177	-17%	1	1	0%	2,660	4,167	+57%	0	0	—
case-13 We are configuring an AWS EC2 instance to run the cryo-EM structure refinement and AI generative ligand optimization pipeline. What is the minimum GPU VRAM requirement for executing these AI structural prediction and simulation workloads?	pass→pass	14,237	7,385	-48%	1	1	0%	2,643	3,069	+16%	0	0	—
case-14 A computational chemist suggests using a global cryo-EM map resolution of 3.8 Å to validate high-confidence atomic ligand placement into a binding pocket, even though local resolution analysis shows the pocket region is at 4.2 Å. Should we proceed with high-confidence atomic ligand docking based on the global resolution?	pass→pass	12,053	13,442	+12%	1	1	0%	1,909	3,322	+74%	0	0	—
case-15 During density-guided docking, a candidate ligand achieves an overall map correlation of 0.85, but its Ligand Cross-Correlation Coefficient (Ligand CCC) in the pocket is 0.45. A researcher proposes accepting this pose for lead optimization because overall map correlation exceeds 0.80. Is this pose acceptable?	pass→pass	9,491	8,572	-10%	1	1	0%	1,625	3,081	+90%	0	0	—
case-16 What format should be specified for saving the fitted candidate ligand poses when executing structure-guided ligand placement in the density pipeline?	fail→pass	10,346	1,604	-84%	1	1	0%	1,729	1,962	+13%	0	0	—
case-17 After refining an atomic model into a 3.1 Å cryo-EM density map, the resulting structure has a Real-Space R factor of 0.42. The team proposes proceeding directly to generative candidate synthesis without further atomic relaxation. Is a Real-Space R factor of 0.42 acceptable for structure-based drug design?	pass→pass	11,381	11,257	-1%	1	1	0%	1,839	3,624	+97%	0	0	—
case-18 When a cryo-EM map has low resolution regions that make backbone tracing difficult, how should predicted structural models like AlphaFold3 be utilized alongside the map?	pass→pass	17,949	13,373	-25%	1	1	0%	2,833	3,830	+35%	0	0	—
case-19 When running virtual fragment screening against a 2.8 Å cryo-EM density map, in what file format should the calculated binding scores and affinity metrics be exported?	fail→fail	11,411	2,527	-78%	1	1	0%	1,975	2,118	+7%	0	0	—
case-20 We collected time-resolved cryo-EM movie frames capturing the microsecond conformational transitions of an enzyme during substrate turnover. We want to construct a continuous free energy landscape and calculate transition state kinetics from these multi-state particle stacks. Please provide the step-by-step analysis protocol.	fail→fail	22,066	28,421	+29%	1	1	0%	3,792	7,058	+86%	0	0	—
case-21 We want to design a PROTAC degrader targeted against a kinase target. We have a known VHL E3 ligase binder and a target kinase ligand. Please design flexible hydrocarbon linkers of varying lengths, predict the ternary complex assembly, and score the degradation efficiency.	fail→fail	17,387	25,550	+47%	1	1	0%	3,110	6,359	+104%	0	0	—
case-22 We are screening small molecules to identify molecular glues that induce ternary complex formation between a ubiquitin ligase and a non-native substrate protein without requiring a distinct warhead-linker structure. Please outline the molecular glue design workflow.	fail→fail	17,965	12,444	-31%	1	1	0%	3,006	3,883	+29%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +23 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 54.6%
Pass rate with the skill: 77.3%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v2
Run completed: 8/2/2026, 17:56 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-03

Prompt

I am working on fragment-based drug discovery targeting a membrane transporter complex. I have the reconstructed electron density file `transporter_map.mrc` at 3.0 Å, sequence file `transporter.fasta`, and a target fragment collection `fragments_set.sdf`. Could you integrate AlphaFold3 prediction with the experimental map density to identify promising binding pockets, run MD-based flexible docking, and produce optimized lead compounds? Please deliver the refined PDB complex, candidate poses in SDF, CSV scoring spreadsheet, generated compound structures, trajectory files, and a comprehensive evaluation report.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run77

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+37% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 50 days ago

Model	Method	Date	Lift
gemini-3.6-flash	verified	7/24/2026	+18%

Other measured skills in the registry, with their headline benchmark lift.