Main Page
From SysbioWiki
Broadly defined, what we do is a mixture of systems biology and synthetic biology. Said another way, we are reinventing how and why biological research is done.
Much of the biological research today is lost in time. Specialists toil for years on a single problem to understand an infinitesimally small corner of this amazing machine we call a living cell. This work is safely hidden away in a dialect known by few, in journals read by few, with little thought about how this work will be integrated with the work of others to change us as a species.
Our aim is to push biological research into the future by making it automated, systematic, and holistic. Biological experiments done with human hands are slow and error prone, biological analysis done with human brains is strikingly narrow sited. Automation of both the laboratory and the experimenters mind frees us up to do the things that humans are really good at—namely be creative, break things, and see how our knowledge can change us.
Following this aim, our group is currently strongly focused on data driven modeling using Bayesian networks. Bayesian networks are a class of tools used in machine learning that can automatically structure complex and noisy datasets into a (sometimes) human interpretable network. This network can be used to make predictions about the future, generate error estimates, and identify locations of low data density or data ambiguity. These models make no assumptions about the functional form of the phenomena, thus can capture linear as well as nonlinear, or even discontinuous phenomena equally well.
The key challenge with Bayesian network analysis is that learning Bayesian networks is incredibly computationally demanding. To get around this problem, the group is pioneering methods to handle topologically constrained networks—networks where we force part of the structure and search out the rest. Many problems in biology can be directly mapped to a topologically constrained Bayesian network problem. Many topologically constrained Bayesian network problems allow vast simplifications that allow them to be computed on only a small supercomputer.
We use our Bayesian network engines to drive intelligent experiments, ranging from drug discovery, biomarker detection, regulatory pathway reconstruction, and synthetic genetic machine development. Some of our experiments are run using automated microfluidic devices that we build in house. Others are run in collaboration with others.
Where are we going?
We constantly searching for something better. We have already adopted tools from coding theory, number theory, combinatorics, many branches of engineering, and biologically inspired circuitry. But this is just the start.
If you know of something better, please let us know! Or work with us! Or work against us (but please publish)!
