vicdyf: Variational Inference of Cell state Dynamics with fluctuation

vicdyf is intended to estimated cell state dynamics with fluctuation from spliced and unspliced transcript abundance.

Instalation

You can install vicdyf using pip command from your shell.

pip install vicdyf

Usage

You need to prepare an AnnData object which includes raw spliced and unspliced counts as layers named as spliced and unspliced like a scvelo data set. Apply vicdyf workflow on the object:

import vicdyf
adata = vicdyf.workflow.estimate_dynamics(adata)

vicdyf.workflow.estimate_dynamics have optional parameters as below:

• use_genes: gene names for dynamics estimation (default: None)
• first_epoch: number of epochs for deriving latent representation (default: 500)
• second_epoch: number of epochs for optimizing dynamics (default: 500)
• param_path: a path where the optimized parameters of vicdyf.modules.VicDyfare stored (default: .vicdyf_opt_pt)
• lr: Learning rate for Adam optimizer of pytorch
• batch_size: Size of mini batches in the optimization procedure
• num_workers: Number of workers in data loader of pytorch
• val_ratio: proportion of validation data set
• test_ratio: proportion of test data set
• model_params: a dictionary which describe the configuration of vicdyf.modules.VicDyf. The keys of the dictionary is as below:
  • z_dim: dimension of latent representation (default 10)
  • enc_z_h_dim: dimension of hidden units in encoder layers (default 50)
  • enc_d_h_dim: dimension of hidden units in dynamics encoder layers (default 50)
  • dec_z_h_dim: dimension of hidden units in encoder layers (default 50)
  • num_enc_z_layers: the layer number of the encoder (default 2)
  • num_enc_z_layers: the layer number of the dynamics encoder (default 2)
  • num_dec_z_layers: the layer number of the decoder (default 2)

Here, the AnnData object acuires sevral elements in layers, obsm, obsp and obs.

• layers:
  • vicdyf_expression: Expected gene expression level
  • vicdyf_mean_velocity: Expected gene expression change
  • vicdyf_velocity: Stochasticaly sampled gene expression change
  • vicdyf_fluctuation: Fluctuation level for each gene
• obsm:
  • X_vicdyf_z: Stochasticaly smapled latent representation
  • X_vicdyf_zl: Expected latent representation
  • X_vicdyf_d: Stochasticaly smapled changes of latent representation
  • X_vicdyf_dl: Expected changes of latent representation
  • X_vicdyf_umap: 2D UMAP embeddings of expected latent representation for visualization
  • X_vicdyf_sdumap: 2D UMAP embeddings of X_vicdyf_d for visualization
  • X_vicdyf_mdumap: 2D UMAP embeddings of X_vicdyf_dl for visualization