Gene regulatory network (GRN) inference is a widely used technique to find drivers of cellular processes. Traditional methods largely focus on discrete cell sub-populations, potentially overlooking dynamic gene interactions across a continuous cellular landscape. We developed a novel GRN inference method called NeighbourNet (NNet), designed to work on individual cells from single cell RNA-seq data. NNet uses principal component regression to capture local gene co-expression patterns within each cell's nearest neighbourhood. By combining this information from external data such as a list of transcription factors and their targets, NNet is able to derive a GRN for each cell. It can also aggregate individual cell networks into meta-networks, highlighting key GRNs at different resolutions.
We applied NNet to a range of single cell data, including early hematopoiesis development and brain tumour. NNet was able to recapitulate known findings in these studies, as well as predict new GRNs. Numerically, NNet has been optimised to efficiently analyse networks comprising of thousands of cells without high-performance computers.