Minjae Chung
BME MS Thesis Defense Presentation
Date: 2026-09-10
Time: 11 AM
Location / Meeting Link: UAW 2100

Committee Members:
Melissa Kemp, May Wang, Facundo Fernandez


Title: Spatiotemporal Metabolomic Profiling For Cell Fate Prediction In Human Stem Cell Differentiation Systems

Abstract:
Stem cell differentiation is commonly evaluated through molecular markers, yet differentiation also depends on how emerging cell populations organize within developing tissues. This thesis develops a spatially resolved framework for examining differentiation across three stem cell systems with distinct architectures and analytical scales. First, seventeen spatial features were used to quantify endodermal and mesodermal organization in doxycycline-inducible GATA6 liver organoids. These analyses showed that doxycycline concentration, transgene copy number, and morphogen signaling influence lineage mixing, clustering, radial distribution, and domain adjacency. Second, matrix-assisted laser desorption/ionization mass spectrometry imaging was coregistered with immunofluorescence microscopy to construct a continuous lipid-derived differentiation index in spontaneously differentiating induced pluripotent stem cell colonies. This index retained intermediate cellular states that were not represented by discrete marker classifications and revealed temporal and spatial organization in differentiation-associated lipid profiles. Finally, the framework was translated to directed neural rosette differentiation to test whether metabolomic and spatial features distinguish NCAM-defined states. Metabolomic profiles carried most of the classification signal, while spatial features provided only a modest improvement but helped identify radially structured prediction errors and candidate metabolic gradients. Performance decreased substantially when entire rosettes were held out, demonstrating that random cell-level validation overestimates generalization across spatially distinct structures. Together, these results show that spatial organization provides biologically informative context for stem cell differentiation, while also emphasizing the need for structure-aware validation, standardized spatial reference frames, and greater biological replication before spatial-metabolomic models can be applied predictively across samples.