Making multi-axis Gaussian graphical models scalable to millions of cells

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Biological data analysis often relies on network inference, including gene regulatory networks and cell-cell interaction networks.
• Existing network inference methods often make an 'independence assumption,' limiting their scalability and applicability to large datasets.
• Current multi-axis methods struggle with datasets exceeding a few thousand cells or genes, hindering analysis of modern single-cell RNA sequencing (scRNA-seq) data.

Purpose of the Study:

• To develop a scalable multi-axis network inference method capable of analyzing large-scale biological datasets.
• To overcome the limitations of existing methods in handling high-dimensional and large-volume scRNA-seq data.
• To enable deeper biological interpretation and discovery from complex network structures.

Main Methods:

• Developed a novel multi-axis method for learning conditional dependency networks.
• The method is implemented as a Python package (GmGM) available on PyPI.
• The methodology scales to process datasets with millions of cells within minutes.

Main Results:

• The new method successfully processed large scRNA-seq datasets, achieving results in minutes that were previously computationally infeasible.
• Applied to neuronal cell development data, the method generated gene networks with improved biological interpretability compared to state-of-the-art methods like hdWGCNA.
• Simultaneously learned cell networks demonstrated advantages over traditional kNN-based clustering, revealing potential roles for long non-coding RNAs in neuronal development.

Conclusions:

• The developed multi-axis method significantly advances the scalability and applicability of network inference for large biological datasets, particularly scRNA-seq.
• This approach facilitates more robust biological interpretation and the discovery of novel molecular mechanisms, such as the role of lncRNAs in development.
• The GmGM package provides a valuable tool for researchers analyzing complex biological networks.