UnionCom

We released new version UnionCom2 with fast and efficient implementation

Paper

Unsupervised Topological Alignment for Single-Cell Multi-Omics Integration

• Corrigendum: In the above paper, $1_{n_x\times n_y}$ on Pages i50 and i51 should be corrected as $1_{n_x\times n_x}$. The authors thank Dr. Chanwoo park from Seoul National University for pointing out this typo.

Enviroment

python >= 3.6

numpy 1.19.5
torch 1.7.0
scipy 1.4.1 torchvision 0.4.1
scikit-learn 0.23.2
umap-learn 0.3.10

Install

UnionCom software is available on the Python package index (PyPI), latest version 0.4.0. To install it using pip, simply type:

pip3 install unioncom

Change Log

v0.4.0

• Add batch effect correction method by setting integration_type="BatchCorrect";
• Add more distances (e.g. cosine, cityblock, see sklearn.metrics.pairwise) to formulate distance matrices.
• Fix some bugs;

Examples (jupyter notebook)

• Integration of simulation 1 in UnionCom paper

• Integration of simulation 2 in UnionCom paper

• Integration of simulations in MMD-MA paper

• Batch correction of HSC data

• Integration of scGEM data

• Integration of scNMT data

Each row should contain the measured values for a single cell, and each column should contain the values of a feature across cells.

>>> from unioncom import UnionCom
>>> import numpy as np
>>> data1 = np.loadtxt("./simu1/domain1.txt")
>>> data2 = np.loadtxt("./simu1/domain2.txt")
>>> type1 = np.loadtxt("./simu1/type1.txt")
>>> type2 = np.loadtxt("./simu1/type2.txt")
>>> type1 = type1.astype(np.int)
>>> type2 = type2.astype(np.int)
>>> uc = UnionCom.UnionCom()
>>> integrated_data = uc.fit_transform(dataset=[data1,data2])
>>> uc.test_LabelTA(integrated_data, [type1,type2])
>>> uc.Visualize([data1,data2], integrated_data, mode='PCA') # without datatype
>>> uc.Visualize([data1,data2], integrated_data, [type1,type2], mode='PCA') # with datatype

Parameters of class UnionCom

The list of parameters is given below:

• epoch_pd: epoch of Prime-dual algorithm (default=2000).
• epoch_DNN: epoch of training Deep Neural Network (default=100).
• epsilon: training rate of data matching matrix F (default=0.01).
• lr: training rate of DNN (default=0.001).
• batch_size: training batch size of DNN (default=100).
• rho: training damping term (default=10).
• delay: delay steps of alpha (default=0).
• beta: trade-off parameter of structure preserving and point matching (default=1).
• perplexity: perplexity of tsne projection (default=30)
• kmax: maximum value of knn when constructing geodesic distance matrix (default=40).
• output_dim: output dimension of integrated data (default=32).

The other parameters include:

• log_pd: log step of Prime Dual (default=1000).
• log_DNN: log step of training DNN (default=10).
• manual_seed: random seed (default=666).
• distance_mode: mode of distance. ['geodesic' (suggested for multimodal integration), 'euclidean'(suggested for batch correction)] (default='geodesic').
• project_mode: mode of project, ['tsne', 'barycentric'] (default='tsne').
• integration_type: "MultiOmics" or "BatchCorrect". "BatchCorrect" needs aligned features. (default='MultiOmics')

Contact via [email protected]