scGeneRythm: Using Neural Networks and Fourier Transformation to Cluster Genes by Time-Frequency Patterns in Single-Cell Data

Abstract

AbstractClustering genes in single-cell RNA sequencing plays a pivotal role in unraveling a plethora of biological processes, from cell differentiation to disease progression and metabolic pathways. Traditional time-domain methods are instrumental in certain analyses, yet they may overlook intricate relationships. For instance, genes that appear distinct in the time domain might exhibit striking similarities in the frequency domain. Recognizing this, we present scGeneRhythm, an innovative deep learning technique that employs Fourier transformation. This approach captures the rich tapestry of gene expression from both the time and frequency domains. When evaluated across a spectrum of single-cell datasets, scGeneRhythm consistently outperforms conventional approaches. The gene clusters it identifies not only demonstrate heightened statistical significance in enriched pathways but also bring to light underlying gene relationships previously obscured. Through integrating frequency-domain data, scGeneRhythm not only refines gene grouping but also uncovers pivotal biological insights, such as nuanced gene rhythmicity. By deploying scGeneRhythm, we foster a richer, multi-dimensional understanding of gene expression dynamics, enriching the potential avenues of cellular and molecular biology research.