Virginia Tech researchers have received a $2.13 million grant from the National Institute of General Medical Sciences to integrate so-called "top-down" and "bottom-up" approaches in computational systems biology.
The team — comprising T.M. Murali, an associate professor of computer science; John Tyson, a professor of biology; and Jean Peccoud, an associate professor at VT's Virginia Bioinformatics Institute — will develop a combined framework that draws from the two approaches and will use it to study cell division in yeast. They plan to release multiple versions of their software over the course of the five-year funding period.
According to the project's grant abstract, the researchers plan to develop algorithms to search databases to enhance models of cellular systems; come up with new principles to test how these improved models match experimental data; and design experiments to validate predictions made in the first two steps.
The goal is to fuse the top-down approach to computational systems biology — which automatically analyzes large-scale datasets for correlations between genes and proteins — with the bottom-up approach — which creates detailed models that can be simulated in silico and validated in the lab.
While these approaches have largely been developed independently, this project aims to combine each method's strengths into a single framework and will create a hypothesis-generation, -testing, and -validation pipeline that will ultimately provide useful information about cell cycles, Murali told BioInform.
To illustrate, he described a theoretical situation in which a researcher might want to know whether a particular protein plays a role in cell division.
“Hypothesis generation involves [identifying] potential paths by which this candidate protein might control the cell cycle,” he said. “The hypothesis-testing step actually adds these interactions to the cell cycle model and then simulates the augmented cell cycle model to see if it better fits the experimental data.”
He continued, “when we test multiple paths [and find] one [that] best improves the fit to the existing experimental data … we can validate it experimentally.”
Murali explained that the top-down approach searches large sets of information from gene expression experiments and protein interaction studies, for example, to identify correlations between genes and proteins and build representations of subcellular networks.
Tools of the trade in this approach typically include gene clustering software and network algorithms to study protein interaction networks and gene regulatory networks.
“The power of that approach is that you are combining many different aspects of information about cells and each one of them sheds light on a slightly different facet of cellular activity,” so researchers can find “biologically interesting patterns” in their data, he said.
However, it’s tough to design experiments using information from top-down approaches because, generally, they are designed solely to integrate data to identify groups of genes that work together, Murali said.
As a result, researchers have to sift through hundreds or thousands of patterns to figure out what genes would be of the most interest experimentally and would provide additional information about cellular activity.
On the other hand, the bottom-up approach explores particular aspects of cellular activity — for instance what pathways are perturbed when a ligand binds to a receptor — and integrates it with information on related experiments published in the literature to create a model.
Murali referred to work done by Tyson that uses information on critical regulatory interactions that occur in cell cycles to create ordinary differential equations, or ODEs, that best describe each interaction.
Next, researchers comb the literature for rate constants and other parameters for each ODE and combine all the data to create in silico simulations.
These simulations could show investigators the cellular impact of knocking out a pair of genes or increasing rates of gene expression. They can cross-check their findings if experimental data on the mutation in question is available in the literature or perform their own validation experiments in the lab and then refine the simulations as needed.
Building the models takes years, however, and is "painstaking," Murali said, because researchers have to mine published resources manually, figure out what interactions are essential to the model, identify the necessary parameters, and in many cases develop their own computational methods to estimate the parameters.
"What we want to do is [bridge] the gap between these two approaches," he said.
The game plan for this project, therefore, is to generate hypotheses from top-down models and then test these hypotheses by integrating them into bottom-up models before moving on to the validation step.
During the first two years of the grant, the researchers plan to define and refine algorithms for making connections between the two approaches. Simultaneously, they plan to improve algorithms used in the bottom-up approach to obtain better simulations of cell cycles.
In subsequent years, Murali and colleagues plan to create and run simulations, test them experimentally, and then refine their algorithms and models iteratively.
If their research is successful, "methods developed in this project should be relevant to the study of any complex cellular system," Murali said, "including the development of cancer and the spread of infectious diseases."
The VT research group is not alone in its endeavor to link these two disparate approaches. Research groups such as the lab of Steve Oliver, a professor of systems biology at the University of Cambridge, are working on fusing top-down and bottom up paradigms.
Oliver's group is working on modeling the yeast cell and is using metabolic control analysis — described as "a conceptual and mathematical formalism that models the relative contributions of individual effectors in a pathway to both the flux through the pathway and the concentrations of individual intermediates within it" — as a theoretical framework.
In the top-down approach, the researchers identify genes encoding proteins with "high flux control coefficients" and use them to build "a coarse-grained model of the eukaryotic cell, as exemplified by yeast."
On the bottom-up side of things, "individual sub-systems are modeled in detail" requiring "that ‘natural’ biological systems be identified and the degree to which they are (or can be) isolated from the rest of yeast’s networks .... determined." The team is using "flux-balance/flux-coupling analyses, combined with both genetics and metabolomics, to define metabolic and other systems."
Have topics you'd like to see covered in BioInform? Contact the editor at uthomas [at] genomeweb [.] com.