FAVA: Functional Associations using Variational Autoencoders

FAVA is a method used to construct protein networks based on omics data such as single-cell RNA sequencing (scRNA-seq) and proteomics. Existing protein networks are often biased towards well-studied proteins, limiting their ability to reveal functions of understudied proteins. FAVA addresses this issue by leveraging omics data that are not influenced by literature bias. Read the documentation.

Data availability

The Combined Network

Installation:

pip install favapy

favapy as Python library

Read the jupyter-notebook: How_to_use_favapy_in_a_notebook

favapy supports both AnnData objects and count/abundance matrices.

Command line interface

Run favapy from the command line as follows:

favapy <path-to-data-file> <path-to-save-output>

Optional parameters:

-t Type of input data ('tsv' or 'csv'). Default value = 'tsv'.

-n The number of interactions in the output file (with both directions, proteinA-proteinB and proteinB-proteinA). Default value = 100000.

-cor Type of correlation method ('pearson' or 'spearman'). Default value = 'pearson'

-c The cut-off on the Correlation scores.The scores can range from 1 (high correlation) to -1 (high anti-correlation). This option overwrites the number of interactions. Default value = None.

-d The dimensions of the intermediate\hidden layer. Default value depends on the input size.

-l The dimensions of the latent space. Default value depends on the size of the hidden layer.

-e The number of epochs. Default value = 50.

-b The  batch size. Default value = 32.

If FAVA is useful for your research, consider citing FAVA BiorXiv.

Other Relevant publications:

The STRING database in 2023.