I received my PhD in algebraic topology from Trinity College, Cambridge in 1981 and was subsequently L.E. Dickson Instructor at the University of Chicago and Research Fellow in Pure Mathematics at Trinity College, Cambridge. I joined Hewlett-Packard (HP) Research in 1987, becoming Director of Basic Research in Europe in 1994. I held visiting academic appointments at Stanford, MIT and Cambridge and was Professeur Invitée at the École Normal Supérieure in 2002. I was appointed to the Council of the Engineering and Physical Sciences Research Council (EPSRC), the UK's equivalent of the NSF, from 1999-2001. I left HP in 2001 to become Visiting Scientist and Head of Systems Biology at the Bauer Center for Genomics Research at Harvard. I joined the Department of Systems Biology in 2002.

Research Summary

Our interest is in signal transduction in eukaryotic (metazoan) cells. Cells use transmembrane receptors to "see" their external environment and somehow transduce these signals into decisions about whether to grow, move, divide, die, hunt, differentiate, ... We now know quite a bit about some of the molecular components involved in this information processing but we lack an understanding of how the system of interacting components produces the phenotypes that we see in the lab. What we mean by "understanding" here is something like what engineers mean when they say they understand how a radio works: why the various components and sub-systems are there; what happens when the system is modified in some way; how to fix it when it breaks. We approach this from several directions.

We are developing an old but under-exploited biophysical technique, Fluorescence Correlation Spectroscopy, that can tell us, in principle, about protein numbers and protein-protein interactions in single cells. We are using this to work out how KSR, the scaffold for the MAP kinase cascade, contributes to information processing downstream of EGF signalling. While this kind of single cell data is essential, it misses the phenotypic variation that is always seen in cell populations, even those composed of genetically identical cells. We use high throughput fluorescence microscopy to measure this variation and to quantify the phenotypic behaviour of cells in response to external signals.

One of the central problems in signal transduction is to get a better grasp of phosphorylation state. Many proteins are phosphorylated, some heavily. A single substrate molecule with n phosphorylation sites has, in principle, 2^n phosphorylation states. These states are regulated through the dynamical interactions of kinases and phosphatases and this is critical to correct information processing. We know very little about the details of this regulation and we are studying this through both theory and experiment.

The data and insights from experiments help us build mathematical models of signal transduction systems. We believe this is crucial to understanding them in the ways described above. Even simple systems can exhibit very non-obvious behaviours. By disentangling the behaviours mathematically, we get a better intuition for how the system might behave and this guides the experiments we do to understand how it does behave.

The problem with mathematical models of biological systems is that they can get very complicated very quickly. Of course, it is much better if we can simplify or abstract these into something manageable, as we do in physics, but we do not understand how to do this systematically in biology. Perhaps we will, eventually, but for the moment you either stick to simplifiable systems or you confront biological complexity and learn to live with it. We take the latter approach.

It is easy to build quite large models "by hand" and that is how nearly everyone does it. It is much harder to modify these models incrementally and even harder still to take advantage of other people's models as parts of your own. We have designed a computational infrastructure that will help everyone build models in an incremental and modular way. We think something like this will be essential if model building is to evolve from an activity carried out by computational biologists to an everyday practice among all biologists. The infrastructure currently consists of a programming language, little b, written in LISP, together with a growing library of descriptions of biological and biochemical entities. We use little b to help us incrementally build models of the systems we study experimentally.