Unlike molecular biology, systems biology study biological systems as a whole. It doesn't agree with the intuition of breaking down systems into smaller parts, studying the parts one-by-one and hoping to be able to reassemble all the parts again at a later time. To be able to see "the bigger picture" of biological systems, which is full of mystery, redundancy and complexity, we need to include as much as possible, "necessary" information. But representations nowadays that describe relationships among molecules, organelles and cells up to physiological systems at the organism level often require excessive computations and resources. Oftentimes, it feels like we're driving a sports car into the city streets where we can't actually drive fast. In this thesis, we present an abstract way to modeling animal embryo development. As an easy reference, we use the drosophila embryo as a model system.
By looking at the amount of proteins and nucleic acids expressed in cells, we can understand and infer how individual cells are organized, becomes diversified and how `architectural' structure of organisms are constructed. A simple way of making two cells different from each other after cell division is to ensure that one cell activates genes different from those of the other. Thus, we refer to the genome which encodes genomic interactions as the genotype and the resulting pattern we see in terms of gene expression as the phenotype.
The first part of our thesis introduces a methodology to describe how patterns in animals are formed in terms of gene expression levels. To do that, we describe gene regulatory networks with a method similar to the work of Bodnar et. Al. (Bodnar1996) where a set of Boolean rules represent inter-gene activity (inhibition or activation) that affect chromatin states of the genes involved. We refer to chromatin states as the protein contents of the cells and its neighboring cells prior to and after cell division. For instance; if a gene A activates (inhibits), a gene B, we model a connection from node A to node B with an arrow (a t-bar). And if gene B influences a gene C and by following the directed connections, we can easily see, that gene A directly affects the chromatin state of gene B and indirectly influences that of gene C. Chromatin states take the value 0 for no expression, 1 for low threshold, 2 for mildly expressed and 3 for a high level of expression. Even with such abstractions, we were able to replicate Bodnar's work and model drosophila embryogenesis pattern formation up to the pair-rule segmentation stage.
The second part of our work answers another question, "Given the pattern, how much can we know about the dynamics like the number of genes or what kind of genes are required to produce it?" Since we consider genetic regulatory networks from a discrete dynamical network perspective and we define the range of stable cell types which exist among identical genes as the separate attractors or basins of attraction into which network dynamics settles from various initial states, we saw trajectories leading to attractors as the pathways of differentiation as earlier described by Kauffman et. Al. (Kauffman1996). And by relating expression trajectories with the pattern and using it as a fitness function, we showed that Genetic Algorithm can be used to inferring or should we say, "reverse engineering", the gene regulatory networks. As we used the drosophila embryogenesis pattern mentioned above as an example, we were able to find a number gene regulatory networks that produce such patterns.
Our work merely presents an infrastructure wherein abstraction of gene expression data or patterns can be used to infer gene regulatory networks. Since our method relies mainly on random search for an optimum individual from a huge search space, proper input of biologically defensible information will prove vital.