CellGrid

Cell classification by learning known phenotypes

Install

$ pip install cellgrid

Get Started

1. Create a schema file.
   Cellgrid trains a set of machine learning models in a hierarchical structure in order to classify the cell populations in the same manner. This schema is defined as a list in a json, in which each element contains:

   • name.
   • parent. Name of the parent model.
   • model_class_name. Name of the The base model class. The following options are supported:
     • xgb
     • random-forest
     • linear-regression
   • markers. The markers that are used for training the model.
   • For example:

   [
       {
           "name": "all-events",
           "parent": null,
           "model_class_name": "random-forest",
           "markers": [
               "Ce140Di",
               "Ir191Di"
           ]
       },
       {
           "name": "cells",
           "parent": "all-events",
           "model_class_name": "xgb",
           "markers": [
               "CD45",
               "HLA-ABC",
               "CD57",
               "CD19",
               "CD5"
           ]
       },
       {
           "name": "CD4T",
           "parent": "cells",
           "model_class_name": "xgb",
           "markers": [
               "CD5",
               "CD4",
               "CD8a",
               "CD31",
               "CD25",
               "CD3e",
               "CD7"
           ]
       }
   ]

2. Train a GridClassifier.

   from cellgrid.preprocessing import transform
   from cellgrid.ensemble import GridSchema, GridClassifier    
   
   #load schema from the json file
   schema = GridSchema.from_json(path_to_schema)
   #transform the data
   x_train = transform(x_train)
   #train the classifier
   clf = GridClassifier(schema)
   clf.fit(x_train, y_train)

3. Score. Return the F1 score of every model.

   x_test = transform(x_test)
   clf.score(x_test, y_test)

4. Predict

   x = transform(x)
   y = clf.predict(x)

5. Save and load

   from cellgrid.ensemble import save_model, load_model
    
   save_model(clf, path)
   clf = load_model(path)

API

GridClassifier

Constructor

GridClassifier(schema)

Arguments

• schema: See above regarding to schema definition.

Methods

fit

fit(x_train, y_train)

Train the classifier

Arguments

• x_train: The single cell dataset.
• y_train: The labels in a hierarchical structure. An example:

layer1	layer2	layer3
cells	B	Naive B
cells	B	IgD+ Memory B
cells	CD4T	Central memory CD4T
non-cells
cells	CD4T

predict

predict(x)

Predict the hierarchical labels for dataset x

score

score(x_test, y_test)

Return F1 scores of every model.

GridSchema

Methods

from_json

from_json(filepath=None)

Load the schema from a json file. See above regarding to schema definition.

License

MIT

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.