Topic

Whole-cell biocomputing, by M. L. Simpson, G. S. Sayler, J. T. Fleming and B. Applegate out of Oak Ridge National Lab, University of Tennessee and Purdue University.

Open Access Journals

This paper is published in Trends in Biotechnology by the scientific megapublisher Elsevier. It is a shame that this article is not available for everyone to read. My comments here, and on my OpenWetware notes page, are open for anyone to read, and I encourage you to add a comment to this post, or edit the notes yourself. By participating in OpenWetWare, you promote the open sharing of scientific knowledge, and send the message to traditional publishers that you promote this cause.

Idea

This is a review article that covers recent progress (up to 2001) in using whole-cell genetic systems as biocomputing platforms. Rather than emphasize the role of one particular genetic subsystem (a single switch for example), the authors promote the view of the complex genetic machinery of the entire cell as a computational system.

The paper draws analogy between the ideas behind silicon-based integrated circuits and modern understanding of biochemical-based genetic circuits. The authors include a discussion of several natural cellular phenomena that can be viewed as complex computations:

1. Directed motility such as chemotaxis, phototaxis and magnetotaxis viewed as an input-reasoning-actuation system.
2. Symbiotic colonization as a complex computational communication system.
3. Group formation (such as biofilm formation) as a complex computational system.

This discussion leads the authors to consider engineering the cells genetic machinery to perform computations. Rather than focus on a particular domain-specific problem, the authors discuss the ’silicon mimetic’ approach to biocomputational engineering, in which the primary aim is to build the analog of modern silicon-based integrated circuits as genetic circuits inside cells.

The keystone of the silicon-based computing architectures is the transistor . A transistor can be viewed as a three terminal device where the transport of signal between two of the terminals is controlled by the third (the gate). Combinations of transistors can be put together to perform logic operations and other control structures, which is indeed what is done to build up the modern day processor.

The authors suggest an biological analogy to the transistor whereby a substrate-enzyme-product circuit is controlled by some effector acting on the enzyme, such as an enzyme inhibitor, or some transcription factor controlling the concentration of the enzyme. Such an analogy can be used to construct simple logic gates such as AND, OR, and XOR, which the authors suggest.

The authors mention several weaknesses of the synthetic mimetic approach, as discussed below, but remain optimistic about further engineering of this paradigm, and discuss further avenues of this work.

Discussion

I think the analogy of cells as naturally performing computations is a rich and productive one to pursue, and in this I think the authors do a good job by describing several common biological phenomenon as ‘computational’. In fact, the authors present a very interesting comparison between the information storage capacity between the DNA-based memory and logic units of cells, and that of modern silicon-based computers, suggesting that even the most modern silicon processing techniques don’t come anywhere near the volume-capacity for information storage and processing that a cell already has.

However, I think the analogy between cellular and silicon-based computing can go too far, and I think the authors have done just that. The authors seem to put all their eggs in the basket of the silicon mimetic approach. Before I discuss why I think this is a bad idea, I must first say a couple of words about why it might be a good idea. There are two main reasons:

1. We know a lot about silicon-based computing. There is a lot of infrastructure devoted to designing digital circuits to perform a slew of computations efficiently. If we could somehow figure out how to manufacture these circuits inside cells, all of the theoretical work on circuit design (that is not hardware-implementation specific), should be able to be ported over relatively quickly. Given the sheer volume of knowledge we have in this area, this is a good goal.

2. All silicon-based circuits are built upon one single idea - the transistor. In turn, transistors can be combined to form simple logic gates, such as AND, XOR, and OR. In turn, as the authors point out, any combinatorial logic function can be implemented as a series of these three gates. The idea is that if you get one thing right, you can build up arbitrary complexity by putting it together in complex ways. We already know how to do that (see 1. right above), so all we have to do is get this transistor thing right, and we are home free.

Unfortunately this is not true, and the authors even admit there are problems. While the problems are many, I’ll list just a few:

1. Digital circuits require low noise levels to have deterministic states. These boolean functions we are talking about implementing take two inputs, and return an output. Each of the inputs and outputs are boolean, meaning that they are one of two things - true or false (or 1 or 0). In silicon you have techniques to manufacture a system that has well defined states defined by the amount of electric charge in a certain region - if there is no charge in the region, the state is false, if there is above a certain threshold of charge, the state is true. Ideally you would have the same thresholding system inside a cell, but instead of charge, you would measure molecular concentration. The problem is, molecular concentrations in cells can fluctuate with much more noise that electric charge in silicon circuits. Noise is rampant in cells. The problem is, if you can’t rely on well-defined inputs, you can’t expect deterministic outputs, and your boolean logic functions turn into some probabilistic system. I don’t know, but you might be able to compute with such a system, but you certainly cannot compute with it in the same way as a deterministic silicon system.

2. Digital circuits are insulated from each other. There has been a lot of effort to cram more and more transistors into a smaller and smaller area of silicon. Much of the work to do this requires better insulation techniques so that an input of true on one transistor does not affect the inputs on its neighboring transistor. A central problem in biocomputing is how to insulate biochemical networks from each other. After all, the cell is just a bag of chemicals. I discussed this problem in my previous journal club about Programmable Cells, and mentioned that if you implement a computational unit as a network of a few molecules, you cannot have multiple copies of that unit in a single cell without the units interfering with each other. In other words, you won’t be able to put together multiple AND’s, XOR’s and OR’s together inside a single cell to make something more complicated, at least with the current biotechnology. However, the authors do suggest using multiple cells in a computation, where the cell wall acts as the insulation. This is a neat idea and should be explored more, but it might mean that traditional silicon-based computational architectures might need to be abandoned in biocomputing. Even if you could pack multiple non-interfering circuits in one cell, it is not clear that the cell has enough excess DNA capacity to contain all these circuits.

3. These simple logic operations are not adequate for computations requiring memory, and syncing to a clock. Many computations in silicon processing architectures require retrieving past computations from memory, and pipelining many computations together that are synchronized by a system clock. The authors mention these inadequacies, and cite relevant literature on work towards these goals.

The analogy of cells as computational systems is fantastic. However, we need to start thinking outside of the box, and start to consider general computational paradigms that are more aligned to what the cell actually does. It would be interesting to try to model a complex genetic regulatory network as a serial array of logic gates, just to compare the two. I suspect that the cells way of doing things is much more efficient in terms of the space required to encode it and the time required to execute the logic of the network. We should be looking to the paradigms of cellular computation for inspiration to make a programmable cellular biocomputing architecture.

As always, please add a comment to continue this discussion, or edit the my OpenWetWare notes yourself.

Links

• My OpenWetWare Bibliography
• OpenWetWare Notes on this poper
• OpenWetWare
• This paper on PubMed
• Trends in Biotechnology
• Elsevier
• Field Effect Transistor (Wikipedia)
• Central Processing Unit (Wikipedia)
• Programmable Cells Journal Club