Eldan Goldenberg's lab notebook

A new direction for my research

In December, I consulted with my advisor and we decided to shelve the research project I had been working on, because it wasn't getting anywhere, and I was getting increasingly nervous about the chances of graduating in a reasonable amount of time. One day I'd like to pick up the evolution-of-learning thread again, but it would either have to be using a different experimental paradigm or significantly more computing power, because experiments that take a week or more to run and then mostly have negative results are simply not practical. For the moment, this left me without a research project (which incidentally has a lot to do with why this site went quiet again) and that was obviously rather worrying in itself.

After making the decision to drop the old project, I took a couple of weeks off so I could put the frustration of being stuck with the old experiments behind me, and then started figuring out what to do next. I'm now well on the way to having a new project outline, and the past couple of weeks have seen me become progresively more optimistic as I put that together. Details are behind the cut; it's not yet as clearly defined as I'd like it to be, but I'm getting there.

Before dropping the old project, I had approached Ilya Shmulevich of the Institute for Systems Biology about a possible side project. I wanted some experience working with biologists, and to have a local advisor with whom I'd be able to meet more often than I can make the trip to Indiana to see my main thesis advisor. Initially I was only looking for a side project to do in parallel with my thesis, but since I dropped the main thesis project the object of the search changed to something that would interest me and both advisors, and to which they could both contribute some expertise. I'm still working out details, but this is the outline that I sent them today:

---

Based on what I've read so far, I think there are quite a few interesting experiments to be done relating to the general question of where the robustness and evolvability of real genetic regulatory networks comes from. For the time being, my thinking on this subject is quite strongly shaped by a book I read several years ago—Brian Goodwin's "How the Leopard Changed its Spots: The Evolution of Complexity"—which is largely based on the idea that the forms which cells (and therefore assemblies of cells and ultimately whole organisms) may take are strongly constrained by the physical properties of the chemicals they are built out of. The argument is that the set of forms we see in biology is not only not coincidental, but it's also constrained by more factors than adaptation alone: it's also an inherent feature of the particular set of buildings block life-forms are made out of that certain forms are likelier, less likely or impossible to evolve.

I think (and what I've read in the past couple of weeks seems to imply) that a similar thing must be true of network structures, such as genetic regulatory networks. We know that in order to survive, cells have to be both robust in the face of environmental noise, and able to adapt to change in their environments; intuitively this ties in with the idea of operating in the critical regime between order and chaos. I've recently read papers showing that the distribution of functions in random boolean networks affects the likelihood of a network behaving in the ordered, chaotic or critical regime, and that the distribution of functions in boolean networks inferred from gene expression data is not random. I think that these two effects must govern each other: the bias we see in the distribution of functions in real biological networks must be a result of both the likelihood of evolution happening upon a given arrangement and the effect of this bias on the overall behaviour of the network.

I think that in order to come up with a set of experiments worth running, I need to learn more about what kinds of boolean network are found in real genetic regulatory networks. Then in general terms what I would like to do construct networks constrained to have similar characteristics, and compare their properties to random boolean networks. I'm interested in the complexity of the constructed networks, as well as their robustness in the face of noise, and the ease of evolving them to exhibit specific arbitrary behaviours.

---

This may at first look like a bigger change in direction than it really is. What this project idea and my previous work have in common is that they both focus on using idealised models to try to understand the general rules that govern organisms' ability to adapt to their environments. It's just that I'm now looking at a lower level of explanation again, having already moved from a BA in Psychology to my previous work with neural networks.

For the next few weeks, I'll be mostly focused on reading papers , with a little tinkering with toy systems to get to grips with how to build the actual models. I'm a little impatient to be doing the 'real' work already, but I do have a lot of background reading to do, and mostly it's just nice to feel like I'm going somewhere again.