OmniClustifyXMBD: Uncover putative cell stateswithin multiple single-cell omics datasets

Abstract

Clustering plays a pivotal role in characterizing cell states in single-cell omics data. Nonetheless, there is a noticeable gap in clustering algorithms tailored for unveiling putative cell states across datasets containing samples with diverse phenotypes. To bridge this gap, we implement an innovative method termed OmniClustifyXMBD, which integrates adaptive signal isolation with cell clustering. The adaptive signal isolation effectively disentangles gene expression variations linked to distinct factors within individual cells. This separation restores cells to their inherent states, free from external influences. Concurrently, a clustering algorithm built upon a deep variational Gaussian mixture model is devised to identify these putative cell states. Experiments showcase the effectiveness of OmniClustifyXMBD in identifying putative cell states while minimizing the influence of various undesired variations, including batch effects and random inter-sample differences. Moreover, OmniClustifyXMBD demonstrates robustness in its results across different clustering parameters.