True single-cell network inference: modelling gene regulation for individual cells
Robrecht Cannoodt1,2,3,*, Wouter Saelens1,4, Katleen De Preter2,3, Yvan Saeys1,4
1Data Mining and Modelling for Biomedicine group, VIB Center for Inflammation Research, Ghent, Belgium; 2Center for Medical Genetics, Ghent University Hospital, Ghent, Belgium; 3Cancer Research Institute Ghent (CRIG), Ghent, Belgium; 4Department of Respiratory Medicine, Ghent University Hospital, Ghent, Belgium
Reverse engineering gene regulatory mechanisms in a biological system remains a challenging task, largely due to the dynamic nature of regulatory interactions which can vary greatly between cells. Recent developments in high-throughput single cell omics are having a large impact on network inference research, as the increased resolution allows to reconstruct regulomes with much greater accuracy. Two studies [1,2] went one step further by first modelling the progression stage of differentiating cells and using the pseudotime as a prior to improve the network inference.
Taking these developments in mind, we explored the opposite approach: to first reconstruct regulatory interactions for individual cells, and subsequently model the dynamic process by comparing single cell regulatory networks. This novel approach can be used to perform de novo differential network analysis of biological systems without the need to predefine cellular states.
Our method uses a modified random forest feature importance score to rank interactions at single cell level. We applied our method on several datasets, one of which contains developing dendritic cells. We ordered the single cell regulatory networks by inferring a trajectory through them, which closely matched the known differentiation stages. Interestingly, we observed differential activity of multiple pathway-like subnetworks along the trajectory. Initial benchmarks show promising results, and thus we are optimistic that this approach will provide novel insights into the regulatory dynamics of many biological systems.
 A. Ocone, et al. Bioinformatics, 2015, 31: i89–i96.
 H. Hatsumoto, et al. bioRxiv, 2016.