scMINER: a mutual information-based framework for identifying hidden drivers from single-cell omics data.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Single-cell omics data presents challenges in identifying transcriptional and signaling drivers due to sparsity and post-translational modifications.
• Hidden drivers with low expression or RNA-protein inconsistencies are difficult to detect using conventional methods.

Purpose of the Study:

• To develop scMINER, a mutual information-based computational framework for unsupervised clustering and cell-type specific inference of intracellular networks.
• To identify hidden drivers and analyze network rewiring from single-cell RNA-seq data.
• To capture nonlinear relationships and infer driver activities.

Main Methods:

• Developed scMINER, a mutual information (MI)-based computational framework.
• Applied scMINER for unsupervised clustering and cell-type specific intracellular network inference.
• Benchmarked scMINER against popular single-cell clustering algorithms.

Main Results:

• scMINER outperforms existing single-cell clustering algorithms, particularly in distinguishing similar cell types.
• scMINER infers activities for over 6,000 transcription and signaling drivers without relying on binding motifs.
• Demonstrated scMINER's ability to uncover hidden drivers and dissect regulon rewiring in immune cell heterogeneity, lineage differentiation, and tissue specification.

Conclusions:

• scMINER is a widely applicable, accurate, reproducible, and scalable method for inferring cellular transcriptional and signaling networks from single-cell RNA-seq data.
• The activity-based approach of scMINER effectively addresses the challenges of sparse single-cell data.
• scMINER software is publicly available for broader research use.