scvis is a python package for dimension reduction of high-dimensional biological data, especially single-cell RNA-sequencing (scRNA-seq) data.

License

scvis is free for academic/non-profit use.

Versions

0.1.0

Installation

To install scvis, please make sure that you have the necessary libraries (below) installed. After that scvis can be installed from terminal:

# In terminal
python setup.py install

Dependencies:

• tensorflow >= 1.1
• PyYAML >= 3.11
• matplotlib >= 1.5.1
• numpy >= 1.11.1
• pandas >= 0.19.1

How to use

After installing scvis, you can use the scvis command.

1, the train function

The train function can be used to learn a probabilistic parametric mapping (the exact directories of the input files should change based on their actual positions in the computer system):

# In terminal
scvis train --data_matrix_file ./data/bipolar_pca100.tsv \
    --out_dir ./output/bipolar \
    --data_label_file ./data/bipolar_label.tsv \
    --verbose_interval 50

• --data_matrix_file: a high-dimensional data matrix with the first row as the column names, in the tab delimited format. Each row represents a data point, e.g., the expression profile of a cell.
• --out_dir: path for output files
• --data_label_file: a one column file (with column header) provides the corresponding cluster information for each data point, just used for coloring scatter plots
• --verbose_interval: the mini-bach interval to show running information

A trained model is saved in the folder ./output/bipolar/model/

In addition to the model file, the low-dimensional embedding and the log-likelihoods are also written to two files in ./output/bipolar, and are shown as two scatter plots colored by the given label information and the log-likelihoods (the log-likelihood files are names as *_log_likelihood.tsv and *_log_likelihood.png).

The different components of the objective function are also saved to a file (*_obj.tsv) and shown in a graph (*_obj.png). If you want to plot intermediate embeddings during optimizations, you can set the flag: --show_plot=True

By default, the data_matrix_file is normalized by the maximum absolute value (--normalize=True).

2, the map function

After learning a probabilistic parametric mapping, the map function can be used to embed new data to an existing embedding:

# In terminal
scvis map --data_matrix_file ./data/retina_pca100_bipolar_normalized.tsv \
    --out_dir ./output/retina \
    --pretrained_model_file ./output/bipolar/model/xxx.ckpt

• --data_matrix_file: a high-dimensional data matrix with the first row as the column names, in tab delimited format
• --out_dir: path for output files
• --pretrained_model_file: a pre-trained scvis model by calling the scvis train, where xxx should be replaced by the checkpoint file prefix in the model folder.

As for calling the train command, this command will also output the likelihood files and the low-dimensional embedding files, but without the model files and the objective function trace file and plots.

The data matrix files for calling both train and map should be normalized similarly, i.e., the parameters used to normalize the training data should be used to normalize the test data (for example, if you do PCA on the training data, the eigenvectors from the training data should be used to transform the test data).

For both train and map, if you want to use your own config file, you can pass the config file as a parameter with flag: --config_file

The default config file is in scvis/config/model_config.yaml, and you can use this file as a template to set parameters.