Quickstart: MonjuDetectHM

!uv pip install "copick[all]>=1.10.0" "cryoet-data-portal"

import copick
from copick.ops.sync import sync_tomograms, sync_picks
import cryoet_data_portal as cdp

# Step 1: Establish connection and name mappings
# =============================================

# Connect to the CryoET Data Portal
client = cdp.Client()

# Retrieve all runs from dataset 10440 (CZII competition dataset)
runs = cdp.Run.find(client, [cdp.Run.dataset_id == 10440])

# Create mapping from portal run IDs to Kaggle-compatible run names
# Copick normally uses run IDs by default because they are unique,
# while run names may not be unique across multiple cryoET data portal datasets.
portal_runs_to_kaggle_runs = {str(r.id): r.name for r in runs}

# Map portal object names to competition object names
# The portal uses scientific names (Gene Ontology Term/UniProtKB accession),
# while the Kaggle competition uses simplified names
portal_objects_to_kaggle_objects = {
    "beta-galactosidase": "beta-galactosidase",
    "cytosolic-ribosome": "ribosome",
    "virus-like-capsid": "virus-like-particle",
    "ferritin-complex": "apo-ferritin",
    "beta-amylase": "beta-amylase",
    "thyroglobulin": "thyroglobulin",
}

# Step 2: Configure source data access
# ====================================

# Create Copick root pointing to CryoET Data Portal dataset
# The '/tmp/overlay' path won't store anything - it's just required by the API
portal_root = copick.from_czcdp_datasets([10440], '/tmp/overlay')

# Extract and rename pickable objects for the target dataset
objects = []
for obj in portal_root.config.pickable_objects:
    if obj.name in portal_objects_to_kaggle_objects:
        # Create a copy with the Kaggle-compatible name
        kaggle_obj = obj.copy()
        kaggle_obj.name = portal_objects_to_kaggle_objects[obj.name]
        objects.append(kaggle_obj)

# Step 3: Create target dataset structure
# =======================================

# Create new Copick project with Kaggle-compatible structure
# This will store the synchronized data locally.
# Update the paths as needed for your environment.
target_root = copick.new_config(
    '/tmp/czcdp_dataset_demo/copick_config.json',  # Configuration file path
    '/tmp/czcdp_dataset_demo/',      # Data storage directory
    pickable_objects=objects         # Object definitions with correct names
)

# Step 4: Sync tomographic data
# =============================

# Copick constructs the tomogram type from data portal metadata. The competition used
# simplified names for tomograms, so we map the portal's processed tomogram type as well.
sync_tomograms(
    portal_root,                                           # Source: CryoET Data Portal
    target_root,                                           # Target: Local Copick project
    source_runs=list(portal_runs_to_kaggle_runs.keys()),   # All available runs
    target_runs=portal_runs_to_kaggle_runs,                # Run name mapping
    voxel_spacings=[10.012],                               # Competition voxel size
    source_tomo_types=["wbp-denoised-denoiset-ctfdeconv"], # Tomogram type to sync
    target_tomo_types={"wbp-denoised-denoiset-ctfdeconv": "denoised"}, # Mapping to simplified name
    log=True,                                              # Show progress
    exist_ok=True,                                         # Allow overwriting
)

# Step 5: Sync annotation data
# ============================

sync_picks(
    portal_root,                                           # Source: CryoET Data Portal
    target_root,                                           # Target: Local Copick project
    source_runs=list(portal_runs_to_kaggle_runs.keys()),   # All available runs
    target_runs=portal_runs_to_kaggle_runs,                # Run name mapping
    source_objects=list(portal_objects_to_kaggle_objects.keys()), # Portal object names
    target_objects=portal_objects_to_kaggle_objects,       # Kaggle object name mapping
    log=True,                                              # Show progress
    exist_ok=True,                                         # Allow overwriting
)

## Setup Environment
# Clone the repository
print('Cloning repository...')
!git clone https://github.com/kobakos/MonjuDetectHM.git
# Install dependencies
!cd MonjuDetectHM && uv sync
!cd MonjuDetectHM && uv pip install -e .

# Reset the directory
%cd /content/MonjuDetectHM/

!gdown --id 1ZWs8VLXvicXQhahVBE72CxVzxD_MIcFo
!tar -zxvf monjudetecthm_mlflow.tar.gz
!rm monjudetecthm_mlflow.tar.gz

# Define the path to the MLflow model
model_path = 'monjudetecthm_mlflow'
copick_config_path = '/tmp/czcdp_dataset_demo/copick_config.json'

import copick
import mlflow
from pathlib import Path
import pandas as pd

# Check if the model path exists
if not Path(model_path).exists():
    print(f"Model not found at {model_path}")
else:
    print(f"Loading model from {model_path}")
    # Load the MLflow model
    model = mlflow.pyfunc.load_model(model_path)

    # Define the path to the copick configuration

    # Load CoPick root
    print(f"Loading CoPick root from {copick_config_path}")
    copick_root = copick.from_file(copick_config_path)

    if not copick_root.runs:
        print("No runs found in the CoPick root. Please check your data configuration.")
    else:
        # Select the first experiment to run inference on
        experiment_id = copick_root.runs[0].name
        print(f"Running inference on experiment: {experiment_id}")

        # Prepare the model input
        model_input = {
            'copick_root': copick_root,
            'experiment_id': experiment_id,
            'voxel_spacing': 10.012,
            'threshold': 0.3
        }

        # Run inference
        # This might take a few minutes
        results = model.predict(model_input)
        print("\ninference results: ", results)

Output:
    Loaded ensemble of 3 models on cuda
    Loading CoPick root from /tmp/czcdp_dataset_demo/copick_config.json
    Running inference on experiment: TS_5_4

    inference results:  {'TS_5_4': {'apo-ferritin': {'points': array([[5476.564  ,  530.636  ,  270.324  ],
        [6037.236  , 3484.176  ,  310.372  ],
        [5386.456  , 2893.468  ,  340.408  ],
        [2312.772  ,  440.528  ,  500.6    ],
        [1241.488  , 1221.464  ,  620.744  ],
        [2573.084  , 2182.616  ,  640.768  ],
        [3023.624  , 2853.42   ,  690.828  ],
        [ 981.176  , 2242.688  ,  720.864  ],
        [1581.896  , 1231.476  ,  760.912  ],
        [4645.568  , 2983.576  ,  770.924  ],
        [1491.788  , 1361.632  ,  780.93604],
        [1651.98   , 1181.416  ,  810.972  ],
        [1461.752  , 1251.5    ,  810.972  ],
        [ 780.93604, 5566.672  ,  851.02   ],
        [2382.856  , 4525.424  ,  891.068  ],
        [1501.8    , 1371.644  ,  901.08   ],
        [1712.052  , 5126.144  ,  951.14   ],
        [3974.764  , 1832.196  ,  981.176  ],
        [5236.276  , 3143.768  , 1021.224  ],
        [1641.968  , 5036.036  , 1021.224  ],
        [1772.124  , 4945.928  , 1061.272  ],
    ...
        [5436.516  , 1011.21204, 1331.5961 ],
        [3173.804  , 2723.264  , 1461.752  ]], dtype=float32), 'confidence': array([4.500285 , 4.1581726, 3.801238 , 4.2265325, 3.8788574, 3.6727734,
        4.156523 , 4.516871 , 4.4099774, 3.5110698, 1.9641333, 3.3476985],
        dtype=float32)}}}

{
    'experiment_id_1': {
        'apo-ferritin': {
            'points': array([[x1, y1, z1], [x2, y2, z2], ...]),
            'confidence': array([c1, c2, ...])
        },
        'beta-galactosidase': {
            'points': array([[x3, y3, z3], ...]),
            'confidence': array([c3, ...])
        },
        ...
    }
}