FAVA

FAVA infers high-quality functional association networks from scRNA-seq and proteomics data by compressing high-dimensional data into a low-dimensional space using variational autoencoders.

Key Features:

• Network inference from unbiased data: FAVA utilizes scRNA-seq and proteomics data that are unaffected by literature bias to infer functional associations, addressing biases that favor well-studied proteins in existing protein networks.
• High-dimensional data compression: FAVA compresses high-dimensional omics data into a low-dimensional space using variational autoencoders.
• Scalability: FAVA can process large datasets with over 0.5 million conditions.
• scverse ecosystem integration: FAVA employs AnnData as its input source and functions within the scverse ecosystem.
• Understudied protein interaction prediction: FAVA has predicted 4210 interactions between 1039 understudied proteins.

Scientific Applications:

• Functional association network inference: FAVA infers networks from high-dimensional omics data with higher accuracy than existing methods across diverse real and simulated datasets.
• Understudied protein function discovery: FAVA uncovers functions of understudied proteins by using data not affected by literature bias.
• Protein interaction analysis: FAVA offers novel perspectives on protein interactions through accurate network inference from scRNA-seq and proteomics data.

Methodology:

FAVA uses variational autoencoders to compress high-dimensional scRNA-seq and proteomics data into a low-dimensional space for functional association network inference.