scMINER: a mutual information-based framework for clustering and hidden driver inference from single-cell transcriptomics data
Qingfei Pan, Liang Ding, Siarhei Hladyshau, Xiangyu Yao, Jiayu Zhou, Lei Yan, Yogesh Dhungana, Hao Shi, Chenxi Qian, Xinran Dong, Chad Burdyshaw, Joao Pedro Veloso, Alireza Khatamian, Zhen Xie, Isabel Risch, Xu Yang, Jiyuan Yang, Xin Huang, Jason Fang, Anuj K. Jain, Arihant Jain, Michael Rusch, Michael Brewer, Junmin Peng, Koon‐Kiu Yan, Hongbo Chi, Jiyang Yu
Abstract
Single-cell transcriptomics data present challenges due to their inherent stochasticity and sparsity, complicating both cell clustering and cell type-specific network inference. To address these challenges, we introduce scMINER (single-cell Mutual Information-based Network Engineering Ranger), an integrative framework for unsupervised cell clustering, transcription factor and signaling protein network inference, and identification of hidden drivers from single-cell transcriptomic data. scMINER demonstrates superior accuracy in cell clustering, outperforming five state-of-the-art algorithms and excelling in distinguishing closely related cell populations. For network inference, scMINER outperforms three established methods, as validated by ATAC-seq and CROP-seq. In particular, it surpasses SCENIC in revealing key transcription factor drivers involved in T cell exhaustion and Treg tissue specification. Moreover, scMINER enables the inference of signaling protein networks and drivers with high accuracy, which presents an advantage in multimodal single cell data analysis. In addition, we establish scMINER Portal, an interactive visualization tool to facilitate exploration of scMINER results.