FAVA: high-quality functional association networks inferred from scRNA-seq and proteomics data

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Protein interaction networks are crucial for understanding biological processes but are often biased towards well-studied proteins.
• Literature bias in protein networks hinders the discovery of functions for understudied proteins.
• Omics data, such as single-cell RNA-seq and proteomics, offer a less biased alternative but present challenges in functional association analysis due to data sparseness and redundancy.

Purpose of the Study:

• To develop a computational method for inferring protein functional associations from high-dimensional omics data.
• To overcome the limitations of traditional protein-protein interaction network construction, particularly the bias towards well-studied proteins.
• To enable the discovery of novel protein interactions and functions for understudied proteins.

Main Methods:

• Developed FAVA (Functional Associations using Variational Autoencoders), a novel method that compresses high-dimensional omics data into a low-dimensional space.
• Utilized variational autoencoders to infer functional associations and protein networks from omics data.
• Integrated FAVA within the scverse ecosystem, using AnnData for data input and handling.

Main Results:

• FAVA infers protein networks with significantly higher accuracy compared to existing methods, validated on diverse real and simulated datasets.
• Successfully processed large-scale omics datasets exceeding 0.5 million conditions.
• Predicted 4210 novel interactions involving 1039 understudied proteins, highlighting FAVA's potential for uncovering new biological insights.

Conclusions:

• FAVA effectively addresses the challenge of data sparseness and redundancy in omics data for functional association analysis.
• The method provides a powerful tool for unbiased exploration of protein interactions and functional roles, particularly for understudied proteins.
• FAVA offers novel perspectives on protein interaction networks and contributes to advancing biological discovery through computational approaches.