GRNRI: Gene Regulatory Network Inference using Unsupervised Graph Neural Network

Abstract

The reconstruction of gene regulatory networks (GRNs) from gene expression data is a challenging problem. GRNs provide insight into the complex regulatory relationships between genes and can help improve our understanding of biological processes. However, current methods for inferring GRNs have limitations in accurately modeling these relationships. In this work, we propose GRNRI: a variational auto-encoder model that learns to infer GRNs from single-cell RNA sequencing (scRNA-seq) data in an unsupervised way. Our model is a modified version of Neural Relational Inference (NRI), a powerful framework for learning relational structure from data. We developed a version of NRI that explicitly models the regulatory relationships between genes using a variational auto-encoder. Results show that GRNRI achieves comparable or better performance on most benchmark datasets compared with state-of-the-art methods. Our work introduces a powerful tool for advancing our understanding of gene regulation and its role in biological processes.