gently-meta: Global Registry for Microscopy Resources

Infrastructure for coordinating multiple gently systems.

A single gently instance makes one microscope intelligent. But science doesn't happen on one instrument. A discovery on a sequencer should be able to mobilize microscopes. An imaging result should be able to trigger computation. The bottleneck in modern science isn't the instruments — it's the coordination between them.

The Human Genome Project understood this. It didn't just coordinate sequencing centers — it connected sequencing to annotation to analysis to biology. The breakthrough was coordination across modalities, not within one. No single lab could have done it alone.

Scientific instrumentation today is where sequencing was before the HGP. Experiments are facility-specific, hard to reproduce, limited by single-instrument thinking. What if every intelligent instrument in the world could discover every other?

That's what gently-meta is: a global registry where autonomous instrument agents find each other, share capabilities, and coordinate across modalities.

Imagine:

• A genomics facility in Cambridge discovers an unexpected gene expression pattern
• gently-meta mobilizes microscopes in Boston, Tokyo, and Heidelberg to validate it across diverse samples and imaging modalities
• Results flow to shared storage, processed by HPC in Frankfurt
• A VLM monitors imaging quality in real-time
• The Model project integrates genomic and imaging data to generate the next hypothesis
• Another facility picks up the follow-up experiment — in a different modality entirely

The discovery-to-validation loop that currently takes months of emails and facility bookings happens in hours. No single lab needs to own every capability. The collective sees more than any individual.

Individual gently instances remain autonomous — gently-meta is a registry they query, not a command layer.

Architecture

How it works

1. Resources register their capabilities with gently-meta on startup — microscopes, HPC, VLMs, genomics facilities
2. Researchers submit experiments to gently-meta's queue
3. Reviewers approve and gently-meta matches requests to capable resources across modalities
4. Resources query gently-meta to discover each other — a microscope finds HPC for deconvolution, a genomics facility finds microscopes for validation
5. Results flow to shared data store, accessible to all registered resources
6. Discoveries in one modality trigger experiments in another — closing the loop between observation and hypothesis

Components

Resource Types

Type	Description	Examples
microscope	Imaging instruments	DiSPIM, confocal, widefield
compute	Processing clusters	Slurm HPC, GPU nodes
vlm	Vision-language models	Local LLaVA, Ollama instance
storage	Data repositories	NAS, object storage
analysis	Analysis services	Segmentation, tracking
genomics	Sequencing and omics	Sequencing facilities, proteomics
robotics	Sample handling	Liquid handlers, plate movers

Phase 1: Infrastructure (Current)

• Resource Registry (gently_meta/microscope_registry.py): Register and discover any resource type
• Experiment Queue (gently_meta/queue.py): Priority-based queue with approval workflow
• Sample Specification (schemas/sample_spec.json): Complete experimental context schema
• Coordination API (gently_meta/api.py): REST endpoints for resource discovery and experiment management
• Notification Service (gently_meta/notifications.py): Email-based workflow coordination

Phase 2: Intelligence (Future)

• Active learning loop integration (Model project)
• Hypothesis generation from experimental results
• Automated experiment routing based on scientific goals
• Multi-resource experiment orchestration

Key Concepts

Sample Specification (Sample_Spec)

The Sample_Spec defines the complete experimental context required to produce reproducible biological observations. It serves as:

• Model Input: Describes conditions that produced experimental data
• Model Output: Specifies conditions for the next experiment
• Communication Protocol: Standardized format between researchers, coordinators, and instruments

See schemas/sample_spec.json for the complete schema.

Biological Context Search

Search experiments by biological properties to find samples of a specific kind:

# Find all HeLa cell experiments
curl "/api/v1/samples/search?cell_line=HeLa"

# Find live-cell time-lapse experiments
curl "/api/v1/samples/search?live_cell=true&has_time_lapse=true"

# Find experiments with GFP or mCherry markers
curl "/api/v1/samples/search?fluorescent_proteins=GFP&fluorescent_proteins=mCherry"

# Complex query via POST
curl -X POST /api/v1/samples/search -H "Content-Type: application/json" -d '{
  "organism": "human",
  "antibody_targets": ["tubulin", "actin"],
  "microscope_type": "confocal"
}'

Searchable fields:

Category	Fields
Biology	cell_line, organism, tissue_type
Genetic	genetic_modifications, fluorescent_proteins
Staining	antibody_targets, fluorophores, nuclear_stain
Treatment	compound_names
Imaging	microscope_type, has_z_stack, has_time_lapse, live_cell
Workflow	status

String fields use partial case-insensitive matching. List fields match if any item overlaps.

Python API:

from gently_meta.queue import ExperimentQueue, BiologicalQuery

queue = ExperimentQueue()

# Search with kwargs
results = queue.find_by_biology(cell_line="HeLa", has_time_lapse=True)

# Search with query object
query = BiologicalQuery(
    organism="human",
    fluorescent_proteins=["GFP"],
    live_cell=True
)
results = queue.find_by_biology(query)

# Get summary for a specific sample
summary = queue.get_sample_summary(request_id)

Experiment Request Lifecycle

SUBMITTED ────► UNDER_REVIEW ────► APPROVED ────► SCHEDULED ────► IN_PROGRESS ────► COMPLETED
                     │
                     └────────────► REJECTED
                     │
                     └────────────► REVISION_REQUESTED

Microscope Capability Advertisement

Each gently instance advertises its capabilities:

{
  "microscope_id": "dispim-001",
  "type": "light_sheet",
  "capabilities": ["3d_imaging", "live_cell", "fast_acquisition"],
  "channels": [405, 488, 561, 640],
  "objectives": [10, 20, 40, 63],
  "special_modes": ["TIRF", "FRAP"]
}

gently-meta matches experiment requirements to available instruments.

Quick Start

Installation

pip install -e .

Run the API Server

python -m gently_meta.api

Submit an Experiment

import requests

response = requests.post('http://localhost:5000/api/v1/experiments', json={
    "sample_spec": {
        "sample_id": "hela_timelapse_001",
        "biological_context": {
            "cell_line": "HeLa",
            "passage_number": 12
        },
        "imaging_parameters": {
            "microscope_type": "widefield",
            "channels": [{"name": "DAPI", "excitation": 405}],
            "time_lapse": {"enabled": True, "interval": 300}
        }
    },
    "requester": {
        "name": "Dr. Jane Smith",
        "email": "jsmith@university.edu",
        "institution": "University of Biology"
    },
    "experiment": {
        "microscope_system": "DiSPIM",
        "scientific_rationale": "Study cell division dynamics...",
        "priority": "high"
    }
})

print(f"Request ID: {response.json()['request_id']}")

API Endpoints

Experiment Management

Method	Endpoint	Description
POST	/api/v1/experiments	Submit new experiment
GET	/api/v1/experiments	List experiments (with filters)
GET	/api/v1/experiments/{id}	Get experiment details
PUT	/api/v1/experiments/{id}/status	Update experiment status

Review Workflow

Method	Endpoint	Description
GET	/api/v1/review/pending	Get pending reviews
POST	/api/v1/review/{id}/approve	Approve experiment
POST	/api/v1/review/{id}/reject	Reject experiment

Resource Registry

Method	Endpoint	Description
GET	/api/v1/microscopes	List available resources
POST	/api/v1/microscopes/register	Register new resource
PUT	/api/v1/microscopes/{id}/status	Update resource status
GET	/api/v1/microscopes/find	Find resources by capability

Sample Search

Method	Endpoint	Description
GET/POST	/api/v1/samples/search	Search by biological context
GET	/api/v1/samples/{id}/summary	Get sample summary

Statistics

Method	Endpoint	Description
GET	/api/v1/stats	Queue statistics
GET	/api/v1/health	Health check

Configuration

Environment Variables

export QUEUE_STORAGE_PATH=/path/to/queue.json
export MICROSCOPE_REGISTRY_PATH=/path/to/microscopes.json
export SMTP_SERVER=smtp.gmail.com
export SMTP_PORT=587
export SMTP_USER=your_email@gmail.com
export SMTP_PASSWORD=your_app_password
export FROM_EMAIL=experiments@lab.org
export PORT=5000

Notification Configuration

Edit notification_config.json to configure reviewers for each microscope system:

{
  "enabled": true,
  "reviewers": {
    "DiSPIM": ["ryan@lab.org"],
    "confocal": ["imaging-team@lab.org"]
  }
}

Integration with Model

gently-meta provides the registry and coordination layer; Model provides the intelligence:

gently-meta (Registry)	Model (Intelligence)
Resource discovery	Active learning loop
Experiment queue	Hypothesis generation
Capability matching	Experiment design
Sample specifications	Representation updating

Model queries gently-meta to find resources, then directs experiments based on scientific goals.

# Find available deconvolution service
hpc = gently_meta.find(capability="deconvolution", status="online")

# Find VLM for quality control
vlm = gently_meta.find(type="vlm", capability="image_qc")

Development

Project Structure

gently-meta/
├── README.md
├── LICENSE
├── pyproject.toml
├── schemas/
│   ├── sample_spec.json          # Sample specification JSON schema
│   ├── experiment_request.json   # Experiment request schema
│   └── microscope_capability.json # Microscope capability schema
├── gently_meta/
│   ├── __init__.py
│   ├── queue.py                  # Experiment queue management
│   ├── api.py                    # REST API server
│   ├── notifications.py          # Notification service
│   └── microscope_registry.py    # Resource registry
└── tests/
    └── test_queue.py

Running Tests

pytest tests/

Open Questions

• Federation across institutions?
• Authentication and authorization model?
• Data sovereignty and privacy constraints?
• Handling network partitions and offline operation?

Contributing

Ideas and contributions welcome. See issues for current discussions.

License

See LICENSE file.