The First q-bio Conference on cellular information processing, Santa Fe, New Mexico started today with an opening lecture by William Bialek. Here is an attempt to provide a brief account of his beautiful talk.
Let’s imagine a biological system to be modeled, says William Bialek, and for which sufficient experimental data are available to determine the relevant model parameters: the model can be “located” within a given region of the parameter space. Bialek asks: should we really be satisfied with the observation that this particular set of parameters fits the experimental data or is the fact that the model is located in this particular region of the parameter space tells us something more fundamental? Given that living organisms are shaped by evolutionary processes, which acts on biological functions (well, at least in part, see Lynch, 2007…), finding an appropriate notion of biological functionality would define some “metric” on the parameter space that would tell us why the living organisms is sitting in this particular region. But how can this concept of functionality be defined and quantified?
As concrete examples, Bialek mentioned the typical sigmoidal intput-output curves observed in neurobiology (eg photoreceptor membrane voltage in function of light stimulus, action potential frequency of visual neurons in function of a stimulus feature like motion) or in molecular biology (for example the expression levels of a target gene in function of transcription factor concentration). Why do these input-output functions have this shape and what are the general rules, independent of underlying molecular mechanisms, that explains where the midpoint of the curve is, its width etc… It appears that one of the keys to this problem is to look at the distribution of inputs received by the system in its “natural context” and analyze how this distribution matches the shape of the output function, something that was initially done by Laughlin (A simple coding procedure enhances a neuron’s information capacity. Z Naturforsch 1981 36c, 910–912). One can then ask the question: “how much information does the output provide about the input?” To do so, the level of resolution with which the output can distinguish between certain number of levels has to be taken into account, given the noise in the system. As it turns out, given the observed distribution of input, imposing maximal information transmission leads precisely to the observed input-output curve. The system seems to have been optimized for transmission of information, which represents a very general constraint. Wiliam Bialek went on with additional examples to show that the input-output curve may in fact, for the same system, depend on the context (the distribution of input can change over time) and to demonstrate how this type of analysis also extends to transcriptional regulation using the Bicoid-Hunchback as a model system.
The talk was followed by a quite animated discussion on how to interpret the observation that these particular systems appear to have optimized information transmission. Is optimization of information transmission the only key? Probably not… But William Bialek’s very insightful lecture illustrated what sort of fundamental and general insight can be gained from the combination of a theoretical approach to a simple biological model system. And the resulting discussion also showed that even if the idealized theory might be wrong, the confrontation between theory and reality of biological systems serves to reveal those aspects of a system that are not sufficiently deeply understood.