scMINER: a mutual information-based framework for clustering and hidden driver inference from single-cell...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell transcriptomics data present challenges like stochasticity and sparsity.
• These complexities hinder accurate cell clustering and cell type-specific network inference.

Purpose of the Study:

• Introduce scMINER (single-cell Mutual Information-based Network Engineering Ranger), an integrative framework.
• Address challenges in unsupervised cell clustering and network inference from single-cell data.
• Enable identification of hidden drivers in transcription factor and signaling protein networks.

Main Methods:

• Developed scMINER, a novel computational framework.
• Utilized mutual information for network engineering.
• Integrated transcription factor and signaling protein network inference.

Main Results:

• scMINER demonstrated superior accuracy in cell clustering, outperforming five state-of-the-art algorithms.
• Outperformed three established methods in network inference, validated by ATAC-seq and CROP-seq.
• Successfully identified key transcription factor drivers in T cell exhaustion and Treg specification.
• Enabled accurate inference of signaling protein networks and drivers, beneficial for multimodal data analysis.

Conclusions:

• scMINER provides a robust solution for analyzing complex single-cell transcriptomic data.
• The framework enhances cell clustering, network inference, and driver identification.
• scMINER Portal offers an interactive tool for exploring results and facilitating biological discovery.