This is an unsupervised model for the integrative analysis of single-cell multiomics data, coded in the deep universal probabilistic program Pyro.
This package relies on PyTorch to run. Please install the correct CUDA version that matches your Operating System on the official website. The exact version of the dependent packages can be found in requirements.txt.
Please install the iPoLNG package using the following command in the command line:
pip install iPoLNG
The main function in this package is iPoLNG.iPoLNG. The input consists of the following:
W: a dictionary with each value being torch.tensor. Each value represents the count matrix for one data modality, with rows being cells and columns being features.
num_topics: the number of topics or latent factors.
alpha_k: The shape parameter for the inverse gamma distribution in the prior. The default value is 1.
integrated_epochs: The number of epochs to run in the stochastic variational inference algorithm for iPoLNG. The default value is 3000.
warmup_epochs: The number of epochs to run in the stochastic variational inference algorithm for PoLNG (as a warm-up step). The default value is 3000.
lr: The learning rate in the stochastic variational inference algorithm. The default value is 0.1. You may adjust it according to the characteristics of the data.
seed: The random seed used in the iPoLNG model. The default value is 42.
verbose: A boolean value. Set to True if you would like to show the learning progress in the model. The default value is True.
Here's the example code for a quick start:
from iPoLNG import iPoLNG
import torch
torch.set_default_tensor_type("torch.cuda.FloatTensor" if torch.cuda.is_available() else "torch.FloatTensor") # enable GPU acceleration if possible
W = iPoLNG.load_example_data()
model = iPoLNG.iPoLNG(W, num_topics=20, integrated_epochs=300, warmup_epochs=500, seed=42, verbose=True)
result = model.Run()
result is a dictionary consisting of the following keys:
L_est: a cell by latent factor matrix, interpreted as the low-dimensional representation of the cells.
Ls_est: a dictionary consisting of several cell by latent factor matries, each interpreted as the low-dimensional representation of the cells in the corresponding individual data modality.
Thetas_est: a dictionary consisting of several latent factor by feature matries, each interpreted as the feature scores in the corresponding individual data modality.
time: the running time for the algorithm (including the warm-up step) in seconds.
loss: a dictionary consisting of the loss (i.e. negative ELBO in the stochastic variational inference algorithm) for individual data modality in the warm-up step and integrated data in the iPoLNG model.
alpha0s: a dictionary consisting of the hyperparameters that control the levels of noise across different data modalities in the iPoLNG model.
Please refer to the vignette for the outputs of the example code.
Zhang W and Lin Z (2023) iPoLNG — An unsupervised model for the integrative analysis of single-cell multiomics data. Front. Genet. 14:998504. doi: 10.3389/fgene.2023.998504
If you encounter any problem when using this package, please feel free to open an issue.