cellflow.model.CellFlow

class cellflow.model.CellFlow(adata, solver='otfm')

CellFlow model for perturbation prediction using Flow Matching and Optimal Transport.

CellFlow builds upon neural optimal transport estimators extending [Tong et al., 2023], [Pooladian et al., 2023], [Eyring et al., 2024], [Klein et al., 2024] which are all based on Flow Matching [Lipman et al., 2022].

Parameters:

• adata (AnnData) – An AnnData object to extract the training data from.

• solver (str) – Solver to use for training. Any name registered in cellflow.solvers.SOLVER_REGISTRY ('otfm' or 'genot' by default, extendable via cellflow.solvers.register_solver()).

Methods

get_condition_embedding(covariate_data[, ...])	Get the embedding of the conditions.
load(filename)	Load a CellFlow model from a saved instance.
predict(adata, covariate_data[, sample_rep, ...])	Predict perturbation responses.
prepare_data(sample_rep, control_key, ...[, ...])	Prepare the dataloader for training from adata.
prepare_model([condition_mode, ...])	Prepare the model for training.
prepare_validation_data(adata, name[, ...])	Prepare the validation data.
save(dir_path[, file_prefix, overwrite])	Save the model.
train(num_iterations[, batch_size, ...])	Train the model.

Attributes

adata	The AnnData object used for training.
condition_mode	The mode of the encoder.
data_manager	The data manager, initialised with cellflow.model.CellFlow.adata.
dataloader	The dataloader used for training.
solver	The solver.
train_data	The training data.
trainer	The trainer used for training.
validation_data	The validation data.
velocity_field	The conditional velocity field.