cellink.tl.external.run_mixmil

cellink.tl.external.run_mixmil(dd, n_pcs=50, donor_key='donor_id', bag_phenotype_key=None, embedding_key='X_pca', likelihood='binomial', n_trials=2, n_epochs=2000, batch_size=64, lr=0.001, encode_sex=True, encode_age=True, additional_covariates=None, dtype='float32')

Train a MixMIL model on donor-level data with flexible covariate encoding.

Parameters:

• dd (DonorData) – DonorData object containing single-cell and donor-level information.

• n_pcs (int, default=50) – Number of principal components to compute if not already present.

• bag_phenotype_key (str) – Column name in dd.G.obs for bag-level phenotype.

• likelihood ({'binomial', 'categorical'}, optional) – Likelihood model for training.

• n_trials (int, optional) – Number of trials if using binomial likelihood.

• n_epochs (int, default=2000) – Number of training epochs.

• batch_size (int, default=64) – Training batch size.

• lr (float, default=1e-3) – Learning rate for optimizer.

• encode_sex (bool, default=True) – Whether to include sex as a covariate.

• encode_age (bool, default=True) – Whether to include age as a covariate.

• additional_covariates (list of str, optional) – List of additional column names in dd.G.obs or dd.G.obsm to include as covariates.

Return type:

Tuple[Dict[str, Any], MixMIL]

Returns:

-results (dict)

Training results from the MixMIL model.

-model (MixMIL)

Trained MixMIL model instance.