03/17/2020 @ 9:30 AM – 10:45 AM | Exhibit Hall – Poster Board No. P377
Developmental Computation with Embryonic Stem Cells
T.B. Knudsen1, T.J. Zurlinden1, N.C. Baker2, and R.M. Spencer3
1US EPA, Research Triangle Park, NC USA
2Leidos and US EPA, Research Triangle Park, NC USA
3General Dynamics and US EPA, Research Triangle Park, NC USA
New approach methodologies (NAMs) that enable in vitro profiling of chemical-biological interactions come with the need for in silico models that translate vast amounts of data and information into toxicological prediction. For fetal systems, these models must reflect the best available knowledge of embryology. A new ToxCast platform using pluripotent embryonic stem cells (H9 line) identified a signal for developmental toxicity potential in 183 of 1065 chemicals (17%) [Zurlinden et al., submitted].
Recursive partitioning using 5-fold cross validation on 80/20 split produced binary classifier models for ToxRefDB developmental toxicity with balanced accuracies of 65% (431 chemicals) to 88% (127 chemicals). PI3K-FOXO signaling was hypothesized as a major determinant of sensitivity.
RNAseq profiles for retinoic acid-induced effects on endoderm-directed stem cells (e.g., FOXA2, SOX17, TBXT, EOMES, LHX1, BMP4) revealed a tipping at 17 nM [Saili et al., 2019]. Here, we expand the performance-based models by a deep-learning strategy that brings embryology of these pathways into the fold using a Compucell3D agent-based model (ES-ABM) to unravel chemical effects on signals / responses for self-renewal (unlimited proliferation), pluripotency (broad differentiation potential), and self-organization (rudimentary anatomical structures).
ES-ABM is initially designed to model stochastic dynamics of locally interacting stem cell agents during retinoid concentration x time exposure, conveying mechanistic hypotheses about key events for chemical-induced alterations of stem cell differentiation and self-organization.
This abstract does not reflect US EPA policy.