Genetic regulation can be characterized using metaphors drawn from the fields of computing and digital electronic circuit design. Even though we can use this analogy, the actual “hardware” (or “wetware”) of cellular logic, chemical reactions in the cytoplasm, is profoundly different from today’s electronic hardware. Here we briefly summarize the organization of genetic regulation from a regulatory circuit architecture perspective and in comparison with electronic logic. We focus on regulation in bacterial cells because bacterial genetic mechanisms are more completely understood today. However, similar organizational principles are thought to govern higher organisms.
Bacterial genetic circuits exhibit hierarchical organization: regulons control groups of operons which control gene groupings. Global regulons coordinate regulation of operons in multiple metabolic pathways, other global regulators act through control of DNA spatial configuration. The biochemical logic in genetic regulatory circuits provides real time regulatory control, implements a branching decision logic, and executes stored programs that guide cellular differentiation extending over many cell generations. In higher organisms the regulatory algorithms may control sequential execution of developmental processes over many years of the organism’s life. This regulatory or adaptive process is related to computing models used in Adaptive Computing Systems or Evolvable Hardware of the type being developed under the current DARPA ACS program. Actual hardware for this program is based for instance on Field Programmable Gate Arrays (FPGAs) that enable the reconfiguration or adaptation of the hardware to the application needs.
The understanding of genetic regulatory circuits can serve as models for innovative adaptive or evolving computing systems or subsystems. The mechanisms that implement bacterial genetic logic functions may be entirely within a single cell, may span many cells, or may function across cell generations. Genetic circuits may cross species boundaries as in symbiotic relations between bacteria and higher organisms. At the DNA level, the abilities of viruses to inject their own DNA into a host, or incorporate host DNS into the viral DNA is a mechanism akin to reading and writing micro- or nano-code for instance in adaptive or evolvable computing hardware. Again, there are synergies with the emerging field of adaptive or evolvable computing, where the traditional distinctions of hardware, runtime software, firmware, and CAD-tool software are going through dramatic changes. Object oriented programming techniques have adapted concepts of inheritance from biology -- without these techniques the recent success in reviving the Mars Rover computer remotely would have not been possible. Object oriented and other programming techniques will become more and more pervasive in software, firmware, and hardware design. They will also become an intrinsic part of compilers and optimizers, either in static, dynamic or “post-mortem” forms.
More sinister, perhaps, are the bacterial mechanisms that co-opt the internal logic of target cells to facilitate penetration or evade defensive responses. Such concepts are certainly reminiscent of software viruses and their future incarnations into software agents and micro UAV's -- another DARPA program. Such developments were the subject of recent U.S. government concerns about the vulnerabilities of critical infrastructures, such as communications, computing, and various distribution networks including power.
At any moment, cellular functions are both implemented by and controlled by the network of chemical reactions involving the collection of molecular species in the cell. In a growing cell, the molecular composition is continuously changing as the cell cycle progresses and the instantaneous regulatory control function also changes continuously. In these networks of interconnected reactions, one regulatory protein can control genes that produce other regulators, that in turn control still other genes so that complex branching and looping networks of interactions are formed. Multi-reagent reactions or genetic mechanisms controlled by multiple input signals are key elements for performing sensor or control-logic functions in these networks. Again, analogies of such processes will first show up in software agents and migrate to reconfigurable or evolvable hardware, the development of which is being greatly encouraged by the availability of the world wide web.
Some operating systems now have automatic software update features -- which brings up the question of reliability and robustness of systems. Again we may want to contrast trusted, highly reliable/available computer systems and todays pervasive cheap but generally unreliable office software. In fairness, embedded computing that is less visible but much more pervasive is geerally more reliable (e.g. the many controllers in cars and infrastructure in general.)
The complement of distinct molecules in the cell and the state of the DNA defines the regulatory logic that establishes how the cell functions at that instant. This logic determines when the cell makes new proteins from DNA-encoded instructions and when existing proteins are destroyed. Such DNA changes can lead to temporary or permanent radical changes in the cell morphology, its active metabolic pathways, and importantly, its responses to environmental signals so that future cellular responses to signals differ from current responses. Thus, the cell’s stored instructions (the genetic material) can be dynamically changed (i.e. reconfigured) according to previously stored instructions. A simple form of such a process is referred to as self-modifying code, the dangers of which are well known. The fact that self-modifying code is "more compact" is the main reason for its use, not surprisingly the same reason viruses use compact (bi-directional) DNA codes.
However, more sophisticated versions of this concept, e.g., in the context of reconfigurable, adaptive, or evolvable systems will appear, with suitable safeguards attached. Security has a number of aspects (privacy, integrity, reliability, availability) that should be built into systems at a fundamental level -- in contrast to today’s systems. Biological systems concepts provide again the lead for adaptive computing and hybrid systems.
Mechanisms that sense conditions inside and outside the
cell are integrated into the regulatory logic so that the cell can adapt
to the needs of the moment. Environmental influences originating both within
and outside of the organism can evoke complex regulatory responses. The
interconnected networks of protein reactions that connect sensors to response
mechanisms are, in a sense, the “nervous system” of unicellular organisms.
The interconnected biochemical elements that can form information processing
circuits in cells have many similarities to neural networks [Bray] .
Biological neural networks have in the past inspired electronic equivalents.
Their use tended to be successful if little or no information was available
from their application domain, and when their metrics are well matched
to the expectations in terms of performance. Biological systems have evolved
to deal with many metrics of success, similarly more recent work on networks
and array architectures [
Flachs]
showed that architectures that can deal with a multiplicity of metrics,
and they outperform simple network architectures that are based on a single
(e.g., least-squares) metric. The appearance of signatures in the
metrics of these advanced architectures is significant, they also exhibit
convergence to global optima, a feature not present in simple artificial
neural network models! This work is now being combined with hierachical
genetic algorithms. Such signatures and the associated affine transformations
(shift, rotate, and scale) are also a hallmark of quantum mechanics.
The rotations can be connected to networks with multiple feedback and vortex
flows; this suggests incorporating rotation or commutator terms in master
equations, e.g. as in the Ginzberg-Landau equation that connects fluid-flow
and superconductivity! Examples of biological network flow models
(strongly & weekly pinned), movies:
The cellular regulatory apparatus includes both short-
and long-term memory mechanisms. The current complement of proteins
in the cell and their physical deployment depends on the cell’s history
and thus is a memory. Long-term memory mechanisms are implemented by more
or less permanent changes in the state of the cell or in the DNA sequence.
In higher organisms, successive cellular state changes during organ differentiation
are largely irreversible. The tight integration of the various forms
of memory in biology is far beyond what today’s electronic memory hierarchies
offer. The design of memory systems is one of the areas where we hope to
learn from biology how to avoid memory bottlenecks with suitable parallelism,
pipelining, and integrated hierarchies. Genetic networks have
many attributes commonly associated with computing. Table 1 shows points
of similarity between genetic logic and the electronic digital logic in
chips in today’s desktop computers. The simplest of genetic circuits is
a regulatory cascade capable of initiating events in sequence, but cells
also use create complex biochemically-based control logic structures. The
capability to create combinatorial controls with feedback when coupled
with memory mechanisms provides every element needed to create a type of
asynchronous sequential logic [McAdams]. Biological regulatory circuits
can have multiparameter combinational control functions at key reaction
cascades that function as “subroutines,” the ability to respond conditionally
to external signals, and the ability to read from stored instructions (the
DNA).
Cascades and parallel paths are also known to be able to reliably turn processes off OR on respectively; however, this doesn't lead to reliable and robust logic. Even two-dimensional arrays of switches that are known to be able to reliably switch processes off AND on, because they still have single points of failure! Control, Estimation, and Information Theory has a mathematical solution to solve the reliable and robustnes problem: non-scalar (matrix or tensor) feedback can be used to narrow error distributions. The question is how to reliably aggregate simple functions to implement reliable and robust larger/global functions. We have discovered such a mechanism involving unitary logic, that in error states enables switches in standby loops (e.g. used in bubble memories). A complex version of this mechanism was proposed in topological quantum computing research. A comparative study on architecture principles has also indicated analogies and even dualities between Particle-Wave interactions in Physics, the "internal model principle" in Control and Estimation Theory, and inter- versus intra-cell interactions. The simplest example we found is the biological cell interaction logic
Bio-Logic: Logical IMPLICATION is the function implemented in this biolpgical mechanism!
From a "hardware implementation" point of view, this suggests switch level pass-gate logic, that supports not only the usual binary (True/False) logic, but also the "high-impedance" or disconnect state that is a logic from of NILL, a higher order logic concept (e.g. used for proving hypotheses in logic.) Strictly speaking, electronics uses this principle as well in bus interfaces. In Physics, this is related to "switchable" interaction Hamiltonians, that help decouple subsystems in a controlled way.
In summary, the detailed implementation and the manner of processing of information in cells is radically different from today’s digital information processing paradigms in today’s computers. However, some of the current research on adaptive and evolvable computing systems has started to take advantage of some of the advanced concepts encapsulated in genetic networks.
In contrast Biology has already implemented advanced computation and communication techniques, a.o. achieving very robust encoding. Even higher order encoding techniques using jump-process models have been observed in biological signal encoding in sensors and decoding in the brain [BruckMorfZeevi].
We are continuing our studies to compare and contrast genetic networks with research in advanced computing systems such as parallel, pipelined, adaptive, and evolvable computing, and encourage synergies between these areas in terms of tool development, analysis, modeling, simulation, and architectural design. We will focus on: Memory Hierarchies, Functional to Structural Mapping, Logic and Control, Adaptive computing, Reconfiguration, Life-Time Issues.
Table 1
electronic logic
Signals electron concentrations
Distribution point to point
(by wires or by electrically encoded addresses)
Organization hierarchical
Digital type Logic, clocked sequential boolean logic
Inherent noise due to
discrete electron events and environmental effects
signal/noise ratio high in most computer circuits
low in encoded communication circuits
Switching speed fast ( >10^6 sec-1)