Cell-Type-Specific Predictive Network Yields Novel Insights into Mouse Embryonic Stem Cell Self-Renewal and Cell Fate

Abstract

Self-renewal, the ability of a stem cell to divide repeatedly while maintaining an undifferentiated state, is a defining characteristic of all stem cells. Here, we clarify the molecular foundations of mouse embryonic stem cell (mESC) self-renewal by applying a proven Bayesian network machine learning approach to integrate high-throughput data for protein function discovery. By focusing on a single stem-cell system, at a specific developmental stage, within the context of well-defined biological processes known to be active in that cell type, we produce a consensus predictive network that reflects biological reality more closely than those made by prior efforts using more generalized, context-independent methods. In addition, we show how machine learning efforts may be misled if the tissue specific role of mammalian proteins is not defined in the training set and circumscribed in the evidential data. For this study, we assembled an extensive compendium of mESC data: ∼2.2 million data points, collected from 60 different studies, under 992 conditions. We then integrated these data into a consensus mESC functional relationship network focused on biological processes associated with embryonic stem cell self-renewal and cell fate determination. Computational evaluations, literature validation, and analyses of predicted functional linkages show that our results are highly accurate and biologically relevant. Our mESC network predicts many novel players involved in self-renewal and serves as the foundation for future pluripotent stem cell studies. This network can be used by stem cell researchers (at http://StemSight.org) to explore hypotheses about gene function in the context of self-renewal and to prioritize genes of interest for experimental validation.

Computational modelling of embryonic stem-cell fate control

“…Over the last decade, microarray and sequencing technologies have been used extensively to measure gene expression in mouse and human ESCs with the aim of identifying master regulators of pluripotency regulation and lineage commitment, and their interactions on a genome-wide scale (Ivanova et al, 2006;Loh et al, 2006;Matoba et al, 2006;Wang et al, 2006;Kim et al, 2008;Ding et al, 2009;Dowell et al, 2013). Although these large-scale studies uncovered an inherent complexity of pluripotency regulation, most attention has been paid to a core unit of three TFs, namely Oct4, Sox2 and Nanog.…”

Novel Insights into Embryonic Stem Cell Self‐Renewal Revealed Through Comparative Human and Mouse Systems Biology Networks

“…For mESCs, we updated our existing compendium of high-throughput mESC data [25] to include work from 62 independent research studies, consisting of ~2.3 million data points from 35 mESC cell lines (all derived from 129S/P/T substrains), spanning 1085 conditions, using 5 different high-throughput data types, and encompassing more than 7 billion gene-pair measurements (Table 1; Supplemental Table S6). …”

“…Learning was achieved by computing the posterior probability of a functional relationship between training set gene pairs given all evidential data [36-39] as previously described (Equation 1) [25]. $$P(FR\mid E_1,E_2,\dots E_n)=\frac{1}{Z}P\left(FR\right)\prod _{i=1}^{n}P(E_i\mid FR)$$ Where FR is a binary hidden variable representing whether a gene pair is functionally related, P(FR=1) is the predicted probability that a pair is functionally related, E i represents the evidence score of the gene pair for the i th dataset, and Z is a normalization factor.…”

“…Naïve Bayes nets impose a strict assumption of independence between evidence data that is likely violated by many of our input datasets, as such, we performed regularization to quantitatively correct for the amount of redundant information contained in each dataset as compared to all other datasets in the compendium (Equations 2 and 3) as previously described [25]. $$S_k=1+H{\left(D_k\right)}^{-1}\sum _{i\ne k}I(D_i;D_k)$$$$P(F{R}_{i,j}\mid E_1,E_2,\dots E_n)=\frac{1}{Z}\prod _{k=1}^n\frac{\alpha P[D_k=d_k\left(g_1{g}_2\right)]+{\alpha }^{S_k}-1}{\alpha +\mid D_k\mid {\alpha }^{S_k-1}}$$ Where S k is a heuristic sum of shared information relative to the dataset's entropy used to weight the strength of prior belief in a uniform distribution for the dataset, H refers to Shannon entropy, and I(D i ;D k ) refers to mutual information.…”

“…To analyze the network topology we calculated the scaled degree ( K i ) for each gene as previously described [25, 42]. To determine genes strongly associated with self-renewal (rather than the general connectivity measured by degree), we identified a set of “core” self-renewal genes from our positive training set for each species as those genes tightly connected to other training set genes as previously described (173 mESC genes; 111 hESC genes) [25].…”