Deep exploration of logical models of cell differentiation in human preimplantation embryos.

The advent of single-cell transcriptomics (scRNA-seq) has provided unprecedented access to specific cell type signatures, including during transient developmental stages. One key expectation is to be able to model gene regulatory networks (GRNs) from the cell-type scRNA-seq signatures. However, most computed GRNs are static models and lack the ability to predict the effects of genetic or environmental perturbations. Here, we focus on the maturation process of the trophectoderm (TE), the outer layer of cells of human embryos, which is critical for their ability to attach to the endometrium. Addressing this challenge required overcoming two major limitations: (i) handling the search space generated by the high dimensionality of single-cell data, (ii) the lack of feasible perturbation data for certain biological systems, which limits validation or generation of dynamic models. To address these challenges, we created SCIBORG, a computational package designed to infer Boolean networks (BNs) of gene regulation by integrating single-cell transcriptomic data with prior knowledge networks. SCIBORG uses logic programming to manage the combinatorial explosion. It learns two distinct BN families for each of the two developmental stages studied (TE and mature TE) by identifying specific gene regulatory mechanisms. The comparison between these two BN families reveals different pathways, identifying potential key genes critical for trophectoderm maturation. In silico validation through cell classification into studied stages reveals balanced precision 67% - 73% for inferred BN families. We demonstrate that SCIBORG is a powerful tool that integrates the diversity between gene expression profiles of cells at two different stages of development in the construction of Boolean models.