SIMS: Scalable, Interpretable Modeling for Single-Cell RNA-Seq Data Classification

SIMS is a pipeline for building interpretable and accurate classifiers for intentifying any target on single-cell rna-seq data. The SIMS model is based on TabNet, a self-attention based model specifically built for large-scale tabular datasets.

SIMS takes in a list of arbitrarily many expression matrices along with their corresponding target variables. The expression matrices may be AnnData objects with format h5ad, or .csv. They must be in the matrix form cell x gene, and NOT gene x cell, since our training samples are the transcriptomes of individual cells.

The data is formated like so:

• All matrices are cell x expression
• All label files contain a common column, known as the class_label, on which to train the model
• datafiles and labelfiles are the absolute paths to the expression matrices and labels, respectively

A call to generate and train the SIMS model looks like the following:

from src.models.lib.lightning_train import generate_trainer 

trainer, model, data = generate_trainer(
    datafiles=['cortical_cells.csv', 'cortical_cells_2.csv', 'external/cortical_cells_3.h5ad'], # Notice we can mix and match file types
    labelfiles=['l1.csv', 'l2.csv', 'l3.csv'],
    class_label='cell_state', # Train to predict cell state!
    batch_size=4,
)

trainer.fit(model, datamodule=data)

This will train a derivation of the TabNet model on the given expression matrices with target variable given by the class_label column in each label file.