ScPectral: Spectrally Clustering HypergraphRepresentations of Transcription Networks to Identify Developmental Pathways

Abstract

Transcription Networks, otherwise known as Gene Regulatory Networks (GRNs), are models of biological systems centred on Transcription Factor (TF) interactions. These models equip experimentalists with a powerful computational tool to predict the effects of different genetic perturbations. GRNs are canonically modelled using a digraph, wherein the arcs indicate activation or repression between each pair of nodes to represent the relationships among the TFs. However, gene regulation is accomplished by groups of TFs working in concert, a biological reality the pairwise model neglects. In addition to the paucity of GRN representations incorporating this known TF biology, a persisting challenge to inference of the networks themselves is in accounting for the latent dynamics of gene interactions. In considering this second point, the advent of single-cell RNA sequencing technologies, provides the high resolution data needed to begin effectively inferring temporally-aware models. Despite this, utilisation of temporally-aware statistical metrics to do so has been limited. In addressing these shortcomings to GRN inference, scPectral is introduced as a method to infer a robust dynamic representation of a common GRN motif, the cascade, in the form of a hypergraph. ScPectral is applied to the identification of developmental pathways for known processes to validate its efficacy. Given scPectrals modest success in finding key constituents of developmental pathways, and its ability to do so in a manner requiring no input or annotation of known biology, through further improvement it may develop to become a technique able to aid experimentalists exploring novel development processes. ScPectral is made available at: https://github.com/Dennis-Bersenev/scPectral.