Idle Control Units and Metabolomics
by TechneSpontaneous Reaction Silencing in Metabolic Optimization
From the article:
Performing numerical optimization in glucose minimal media (Materials and Methods), we find that the number of active reactions in such optimal states is reduced by 21%–50% compared to typical non-optimal states, as indicated in the middle bars of Figure 2. Interestingly, the absolute number of active reactions under maximum growth is, 300 and appears to be fairly independent of the organism and network size for the cases analyzed. We observe that the minimum number of reactions required merely to sustain positive growth [7,8] is only a few reactions smaller than the number of reactions used in growth-maximizing states (bottom bars, Figure 2). This implies that surprisingly small metabolic adjustment or genetic modification is sufficient for an optimally growing organism to stop growing completely, which reveals a robust-yet-subtle tendency in cellular metabolism: while the large number of inactive reactions offers tremendous mutational and environmental robustness Papp:2004dn, the system is very sensitive if limited only to the set of reactions optimally active. The flip side of this prediction is that significant increase in growth can result from very few mutations, as observed recently in adaptive evolution experiments.
Reaction irreversibility and spontaneous cascading (article) of inactivity are described as built-in mechanisms that mediate these metabolic adjustments. The authors also point out that 638 out of the 931 reactions in the E. coli glucose metabolic network can be removed whilst maintaining a maximum growth rate in glucose. The mutational robustness as a result of inactive reactions under maximum growth thus act as a sort of preadaptation whereby different pathways can be spontaneously activated under shifting environmental conditions.
The tremendous robustness of these systems raises an interesting question regarding the origins of these non-essential pathways under maximum growth rates. The authors provide a testable hypothesis:
An alternative explanation would be that in variable environments, which is a natural selective pressure likely to be more important than considered in standard laboratory experiments, the apparently dispensable pathways may play a significant role in suboptimal states induced by environmental changes. This alternative is based on the hypothesis that latent pathways provide intermediate states necessary to facilitate adaptation, therefore conferring competitive advantage even if the pathways are not active in any single fixed environmental condition.
This alternative is based on the hypothesis that latent pathways provide intermediate states necessary to facilitate adaptation, therefore conferring competitive advantage even if the pathways are not active in any single fixed environmental condition. In light of our results, this hypothesis can be tested experimentally in medium-perturbation assays by measuring the change in growth or other phenotype caused by deleting the predicted latent pathways in advance to a medium change.
Even more intriguing is the fact that metabolic adjustments are also controlled by anticipatory transcriptional reprogramming in response to environmental changes. It is posited to be as result an “associative learning” paradigm.
Looking at the motor industry again, anticipatory systems and structures have been designed in order to optimize the structure stiffness for a particular crash scenario. Pre-crash sensing is used to adjust structural stiffness and crumple zones in response to a particular deceleration scenario in order to maximize the crash worthiness of the vehicle. It seems this kind of anticipatory programming is an ancient invention, a few billion years old.
Using this information, another core element can be added to an initial state: Robust metabolic networks with tremendous adaptability that "idle" under maximum growth conditions.
Figure 1: An initial state. Reverse engineer ubiquitous core components of various life forms at present. Will it repeatedly produce similar endpoints after evolutionary processes, irrespective of its origin?Ubiquitous components of the system would include the following:
A) A genetic code that is optimized for random searches.
B) Quality control systems (DNA repair, protein quality, programmed cell death).
C) Variation inducers (Cytosine deaminases, Low vs High fidelity polymerases, gene conversion and homologous recombination).
And now tremendously robust metabolic networks that facilitate adaptation.
Does the system react passively whereby random mutations (and other mechanisms) introduce variety for no reason on which natural selection blindly acts upon with regards to fitness? Or does it harness random variation and selection to maintain homeostasis? Are these active entities that search random space for solutions during times of selection pressure with intrinsic quality control systems that act as selection mechanisms to constrain the random search and thus bias the output of a random search?
December 6th, 2008 at 12:51 pm
The cited study shows how genetic models of historical correlations in a multidimensional environment can evolve through random mutation and selection. The genome learns from experience. In other words, evolutionary processes are intrinsically 'intelligent' in a way having nothing to do with some originating teleology.
Evolution occurs within the context of heredity: Heredity is history.
Comment by Zachriel — December 6, 2008 @ 12:51 pm
December 6th, 2008 at 4:04 pm
This claim would be more plausible when backed by evidence that minimally functional genomes evolve from something antecedent. That something needs both specification and demonstration.
Comment by Bradford — December 6, 2008 @ 4:04 pm
December 6th, 2008 at 6:43 pm
The claim doesn't require knowing the origin of the evolutionary process. The process itself can 'learn' regardless of its source. Just like we don't have to know the origin of the Solar System to understand that planets orbit according to the laws of gravity and motion.
Comment by Zachriel — December 6, 2008 @ 6:43 pm
December 6th, 2008 at 6:59 pm
Zachriel:
From an ID perspective an evolutionary "learning process" would fit in with the rest of the narrative thus stengthening an explanation that can begin with the universe itself.
Comment by Bradford — December 6, 2008 @ 6:59 pm
December 7th, 2008 at 10:50 am
Evolutionary processes are intrinsically intelligent because of quality control systems, variation inducing mechanisms, a genetic code reasonably optimal for evolutionary processes and various other robust systems. The genome, it seems, learns from experience and harnesses random variation and selection in order to maintain homeostasis. The article cited demonstrated this beautifully. However, I would like to correlate systems biology (gene expression, enzyme activity, metabolomics etc.) and sequence variability over time in order to see and understand the whole mechanism in action.
Biology is the study of the present and living. The present should give an indication of what happened in the past (history), albeit in limited terms don't you agree?
Comment by Techne — December 7, 2008 @ 10:50 am
December 7th, 2008 at 1:35 pm
Predictive Behavior Within Microbial Genetic Networks simulated evolution with *random mutation*, the central dogma, and a multi-dimensional environment. They showed that such a system could learn, unlearn, and relearn.
And past life. A fossil is evidence of a once living organism. The Theory of Evolution is considered a biological theory.
Yes, that is the implication of what I just said. Past is prologue.
Comment by Zachriel — December 7, 2008 @ 1:35 pm
December 7th, 2008 at 2:00 pm
They simulated it around the "central dogma":
Gene expression and control thereof → Transcription and control thereof → Translation and control thereof, and introduction of variation in the system.
The simulation was based on what we observe in cells: Entities capable of harnessing random variation and selection to maintain homeostasis. The simulation, like cells, was intrinsically intelligent because it had the necessary systems (like cells) to perform/simulate evolutionary processes.
Comment by Techne — December 7, 2008 @ 2:00 pm
December 7th, 2008 at 3:14 pm
Yes, that's correct. It assumes the central dogma. That isolates the genome from the environment. Even then, it is capable of learning.
The only necessity for learning in an evolutionary algorithm is reproduction and selection (assuming a suitably structured environment, e.g. exhibiting locality).
Comment by Zachriel — December 7, 2008 @ 3:14 pm
December 7th, 2008 at 3:29 pm
See, even traditionally viewed simple cells outsmart our best efforts at AI. Learning, perception, manipulation etc. Successful evolutionary processes depend on intelligent systems.
The self-replicating systems and how they are able to control variability (variation induction, quality control systems, genetic code optimal for evolutionary processes, robust metabolic networks etc.) on their own seems to be the necessity for successful evolutionary processes.
Comment by Techne — December 7, 2008 @ 3:29 pm
December 7th, 2008 at 4:07 pm
That is incorrect. It is very easy to show that simple evolutionary algorithms are capable of at least simple learning assuming a reasonably structured environment. (No algorithm, of course, is any better than any other algorithm traversing random landscapes.) We're not talking about biology at this point, only the mathematics of the evolutionary process itself.
I'm not sure how all those apply to the general case, but "quality control systems" seems to be just the inverse of mutational rate.
Comment by Zachriel — December 7, 2008 @ 4:07 pm
December 7th, 2008 at 4:20 pm
What is not intelligent about even simple evolutionary algorithms? Without them, there is no evolution, or even a simulation of evolution.
They remove (or allow/induce) mutations without natural selection. It is part of the self-replication system.
Comment by Techne — December 7, 2008 @ 4:20 pm
December 7th, 2008 at 4:57 pm
As I said, evolutionary algorithms are capable of learning, so we can say they exhibit 'intelligence'.
What you probably mean is that evolutionary algorithms must have a telic origin. But you are arguing in a circle. You can point to learning behavior in evolutionary algorithms, elaborate on it endlessly, marvel at the intricacies, but this is true regardless of the origin! It's inherent to the process.
Just because we model a natural system with a computer doesn't mean that the natural system is thereby telic. A weather simulation models the individual forces that constitute weather. A model of orbital dynamics models the inertia, spin and gravitational attraction of various bodies. An evolutionary simulation models the forces that constitute evolution. The act of creating the model doesn't endow the object of the simulation with teleology.
Yes, if the mutation is eliminated before selection, then it just lowers the effective mutation rate.
Comment by Zachriel — December 7, 2008 @ 4:57 pm
December 7th, 2008 at 5:55 pm
True, but we are able to study models in depth not heretofore possible. One might also ask on what would a study focus to determine whether a system has telic origins. Evolutionary constraints could be fruitful. The evolutionary role constraints play in shaping change from stasis to adaptive radiation is unclear. This is at least partly attributable to an incomplete analyses of potential examples of constraints. A bioinformatic integrative approach to differing levels of organization might be what is needed. Such studies have occurred of course but this looks like a means of identifying telic causes.
Comment by Bradford — December 7, 2008 @ 5:55 pm
December 8th, 2008 at 11:17 am
And, as I said, successful evolutionary processes depend on intelligent systems. Learning is a subset of intelligence and like you said (and as described in the linked article), evolutionary algorithms are capable of learning.
What I mean is: Without evolutionary algorithms, there is no evolution, irrespective of its origin. That is all. You inserted "telic origin". I am not discounting it, I am also not saying it exists because of a telic origin. The effect is there, let's work with that.
A natural system (like a weather system or orbital dynamics) does not self-replicate itself with accuracy so that its progeny can also replicate. A weather system does not have a code that controls information, or quality control systems to bias the outcome of random events. The difference between natural self-replicating systems (like cells) and other natural systems is that self-replicating systems control their own natural processes and bias random events. For example, quality control mechanisms (like DNA repair) actively remove mutations that are randomly induced. Evolutionary algorithms simulate how self-replicating systems evolve and learn.
The forces that constitute evolution are dependent on the nature of the self-replicating entities. In order to simulate evolution of life, you need self-replicating entities. The act of creating the model is to simulate a process, that is it. The process is there, let's work with that.
There you have it, mutations can be randomly induced or they can happen without being induced. They can also be removed irrespective of natural selection. And then, variation can be induced or constrained without mutation (e.g. epigenetic changes).
Comment by Techne — December 8, 2008 @ 11:17 am
December 8th, 2008 at 11:51 am
That doesn't seem to be a well-formed idea. We are using the term 'intelligent' with respect to the capabilities of evolutionary algorithms to learn, but that doesn't mean the individual components are intelligent. Indeed, we can *demonstrate* that very simple replicators without complex controls are capable of learning. All it takes is a population of replicators in a suitable environment.
By definition.
Okay.
Imperfect replicators replicate imperfectly.
Yes, highly-derived biological organisms, the end product of billions of years of evolution, have various control mechanisms. Are you trying to make a point?
Comment by Zachriel — December 8, 2008 @ 11:51 am
December 8th, 2008 at 2:15 pm
Well, even though there is no universally accepted definition of intelligence, systems in traditionally viewed simple cells include perception, learning and manipulation. And successful evolutionary processes depend on those systems.
They simulated the central dogma. Hardly the simple self-replicating Ghadiri protein or self-replicating hammerhead ribozymes.
I prefer to view cells as self-replicators that are capable of controlling variation to adapt to an environment. Perfect replication is not a robust system for adaptation.
Actually, bacteria and other primitive organisms have equally impressive control systems that remove mutations irrespective of natural selection. Successful evolution depend on them, e.g. DNAj.
Comment by Techne — December 8, 2008 @ 2:15 pm
December 8th, 2008 at 2:28 pm
Nevertheless, we can demonstrate learning in evolutionary algorithms without those controls. All we need is a population of replicators in a suitable environment.
Perhaps, but even simple evolutionary algorithms are capable of learning.
As I said, highly-derived biological organisms, the end product of billions of years of evolution, have various control mechanisms. Are you trying to make a point? Because I still don't see it.
Comment by Zachriel — December 8, 2008 @ 2:28 pm
December 8th, 2008 at 2:41 pm
You are discussing the example linked in the OP post aren't you? Simulating the central dogma? Like I said, let's work with that, the central dogma… present in extant living organisms.
Point being, protein folding (such DNAj) quality control mechanisms are present in eubacteria (not billions of years of evolution). The epigenetic example above I saw today and thought it might be interesting to discuss, irrespective of its origins. The quality control mechanisms I was refering to was protein folding, DNA damage control and unicellular programmed cell death. Apologies, I should have been clearer.
Comment by Techne — December 8, 2008 @ 2:41 pm
December 8th, 2008 at 3:03 pm
Evolutionary algorithms generally. Simple replicators can learn.
Primitive cells can be dated to ~3¾ billion years ago. The most recent common ancestor of bacteria and archaea lived ~2½ billion years ago, so by that point they apparently had complex metabolisms. We can grant the point that many cellular control mechanisms are very ancient. So? (I'm not trying to be difficult, but just trying to see if you are saying something obliquely.)
Comment by Zachriel — December 8, 2008 @ 3:03 pm
December 8th, 2008 at 3:21 pm
The central dogma and the evolutionary algorithms modeled on it are hardly simple. Or are you alluding to another example not discussed here?
Irrespective of their origins, these mechanisms are able to constrain evolution without natural selection by removing variation as well as viable mutations from a population. Other mechanisms induce mutations in response to stimuli (cytosine deaminases). So lets work from this state (irrespective of origins) and see if it is able to constrain evolutionary trajectories.
Comment by Techne — December 8, 2008 @ 3:21 pm
December 8th, 2008 at 3:42 pm
The replicator can be genome and phenome. These types of systems are very easy to model, and are quite adept at finding peaks on various types of landscapes. And molecular replicators are subject to artificial evolution in labs, such as at the Whitehead Institute.
It *does* presuppose a vital mechanism—replication! But once we have that, then evolution can proceed apace.
And some areas of the genome are more protected than others. And the Theory of Evolution helps predict which areas.
Evolutionary trajectories are highly constrained, first and foremost by working with what is available.
Comment by Zachriel — December 8, 2008 @ 3:42 pm
December 8th, 2008 at 11:09 pm
Zachriel:
The paper in question states:
In fact, this is a great post, Techne, much like my post on modularity . Life is built around an architecture and this is what evolution tinkers with.
Comment by Guts — December 8, 2008 @ 11:09 pm
December 8th, 2008 at 11:25 pm
I don't think anyone has questioned that biological evolution is much more complex than simple algorithmic evolution. However, even simple evolutionary algorithms are capable of learning. Just pointing to complex integration in biology is not sufficient to determine the origin of that complex integration. It could have been built up in stages, just like the rest of biology—and there is reason to believe it was.
Comment by Zachriel — December 8, 2008 @ 11:25 pm
December 8th, 2008 at 11:50 pm
Sure but I don't really see the point of mentioning this. Very little is certain where the variables are unknown. We entertain hypotheses and follow suspicions.
Comment by Guts — December 8, 2008 @ 11:50 pm