In the post-genomic era, there is a growing need for new experimental paradigms for investigating the links between genomics and biology. While entire genome sequences of most model systems are now available, the task of deciphering the genetic code requires characterizing the phenome of these systems in order to establish the genotype-phenotype links. This need has lead to the development of new quantitative phenotyping technologies across different levels of the biological hierarchy.
In this thesis, I present new computational techniques to conduct image-driven in vivo phenotyping at the cellular level. The techniques have been developed in the context of investigating morphological variations of cells in cancer. Recent findings in cancer biology have provided increasing evidence that the normal cells and molecules that surround tumor cells - collectively termed the tumor microenvironment - are involved in the initiation, growth, and spread of tumors. While examples of this phenomenon have been characterized in studies from a genetic standpoint, the lack of appropriate methodologies have precluded quantitative phenotyping studies at the cellular level. The present work addresses this unmet need.
Based on a novel method that uses local metric-learning to integrate different cellular features, I present a framework to identify major cell types in the microenvironment. I further propose a method to generate phenotypic profiles of cell populations and use the technique to detect the subtle global-level changes that occur among certain cells in the microenvironment in gene knock-out experiments that seek to recapitulate human breast cancer. For supporting the larger scope of investigations into the microenvironment, tools for image analysis, visualization, data management and data analysis have been developed.
By proposing new computational methods for cellular-level analysis, and using them to investigate the tumor microenvironment, I demonstrate that image-driven computational phenotyping provides a viable experimental paradigm to investigate the phenomic aspects of complex processes such as cancer.