Fluorescence microscopy is a primary tool in cell biology. The advancement of microscopy techniques allows for imaging of living cells with higher resolutions, high speed, and high throughput, which motivates researchers to develop novel algorithms to handle these images and answer more fundamental questions. While still useful, approaches to analyze biological images by defining classification or regression problems with efforts to extract features to predict cell states or subcellular localization is insufficient to describe what a cell looks like. Generative modeling methods have been developed intending to comprehensively and accurately describe cell shape and cellular structures. Cell shapes provide both geometry for, and a reflection of, cell function. Various methods for cell shape modeling have been developed, but previous evaluation of these methods in terms of model accuracy has been limited. We compared several methods to build generative models for cell shapes in terms of the accuracy with which shapes can be reconstructed from models. For 3D shape modeling, we developed an improved method based on spherical harmonic analysis (SPHARM) that outperforms existing methods. Then, we integrated this model into the open-source software package CellOrganizer for broader applications. Moreover, we applied this method to model the differentiation process of PC12 cells, a model system to study neuronal differentiation. We showed that the shape model could reconstruct shapes accurately, predict mitochondria distributions, and generate realistic shape trajectories during the differentiation process.

Previous efforts to model signaling and regulatory networks in cells have largely not considered the spatial organization or have used compartmental models with minimal spatial resolution. Fluorescence microscopy provides the ability to monitor the spatiotemporal distributions of many molecules during signaling events. In extending our previous work on modeling protein spatiotemporal distributions for T cell signaling, we developed novel image-based causality inference methods to infer how changes in concentration of a protein in one cell region influences concentrations of other proteins in other regions. The methods identified both known and undiscovered interesting relationships. Moreover, we improved the protein distribution modeling pipeline with alignment refinement, SPHARM-based morphing and computational cell couple annotation methods for more robust, efficient and scalable modeling, with hope for large-scale protein network analysis.

154 pages

Thesis Committee:
Robert F. Murphy (Chair)
James Faeder
Christoph Wülfing
Min Xu

Robert F. Murphy, Head, Computational Biology Department
Tom M. Mitchell, Interim Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu