|
Author
|
Topic: RNA, Scale free networks, degeneracy and evolution
|
Frances
Member
Member # 169
|
posted 12. March 2003 01:04
I would like to use this thread to explore some concepts which may allow us to explore how ID and/or evolution can help us understand many of the observed features in nature. I would for the moment like to exclude front loading from the equation since front loading and methodological naturalism are for all practical purposes indistinguishable.
RNA
Peter Schuster and his colleagues have looked at the networks formed by RNA secondary structures. In their experiments they showed some of the interesting features and consequences of the phenotype genotype mapping in RNA, they showed that like for proteins, there are few common structures which seem to span all of sequence space and which are well connected versus many uncommon structures which often tend to be less well connected. So what relevance does these observations have to evolution? Well, the genotype phenotype mapping and the presence of large neutral networks suggest that neutral evolution plays an important part during periods of stasis followed by periods of innovation when mutations reach new secondary structures. These characteristics may explain why evolution has been so succesful.
Some of their relevant findings are that
1. More sequences than structures
2. Few common and many rare structures.
3. Common structures are found almost everywhere in sequence space.
4. Neutral networks of common structures extend over whole sequence space.
quote:
The existence of extended and connected neutral networks in RNA sequence space was proven by an elegant experiment recently published by Erik Schultes and David Bartel [60]. They designed an RNA sequence which forms two known structures (of chain length (l = 88) with different catalytic activities, an RNA ligase evolved in the laboratory [61] and a natural cleavage ribozyme isolated from hepatitis delta virus RNA [62]. The two structures have no base pair in common. Folding the synthesized chimeric sequence into
structures yielded indeed both activities, although they were substantially weaker than those of the reference ribozymes, the ligase and the cleavage ribozyme, respectively. Only two or three selected point mutations or base pair exchanges are required, however, to reach full catalytic e±ciency. Still, the two optimzed RNA molecules have a Hamming distance around forty from their reference sequences. Then, Schultes and Bartel [60] explored further the mutational neighborhoods and found neutral paths of Hamming distance about 40, by preparing and analyzing series of RNA sequences, in which neighboring sequences differ in a single base or base pair only. Without interruption these neutral paths lead from the RNA with both catalytic activities
to the two reference ribozymes.
I my next postings I will address the scale free networks, their relevance to robustness and degeneracy. Unlike engineered systems which tend to be redundant, biological systems seem to exhibit degeneracy. I will show how evolutionary models can help us understand many of the observations about evolution.
IP: Logged
|
|
Mike Gene
Member
Member # 149
|
posted 12. March 2003 09:56
Frances: I would for the moment like to exclude front loading from the equation since front loading and methodological naturalism are for all practical purposes indistinguishable.
Yes, as I have explained, front loading does not necessarily entail supernatural intervention. Thus, since you are not interested in the perspective of front loading, I'll withhold comment. But I would suggest that you provide the actual references. Thanks.
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 12. March 2003 12:20
Mike, I did not say that I am not interested in the perspective of front loading, I stated that the front loading version of ID (supernatural or 'natural') for all practical purposes is indistinguishable from methodological naturalism.
IP: Logged
|
|
Mike Gene
Member
Member # 149
|
posted 12. March 2003 14:21
Okay, but could you please provide the reference. This looks like stuff I can build on. Thanks.
IP: Logged
|
|
Cre8ionist
Member
Member # 140
|
posted 12. March 2003 15:54
Not to be impatient but
here's the link!
I for one am certainly as interested in Mike's view as I am that of Frances', and my view isn't
really that close to either.
Scanned the doc (it'll take awhile to read) and noticed at the bottom what may be a good thing to keep in mind here
quote:
From analyzing and modelling prokaryotic evolution to an understanding
of multicellular organisms is still a very long way to go and we do not know
yet whether or not generic properties of simple genotype-phenotype mappings
can be used as models. Progress, however, is fast and molecular genetics of
development is witnessing a true explosion of new insights into the enormous
complexity of higher organisms. At the current state of our knowledge it seems
certainly premature to generalize the concepts developed here to evolution of
plants and animals but several features of organismic evolution are already
reflected by simple molecular systems.
............................................................Cre8
IP: Logged
|
|
Mike Gene
Member
Member # 149
|
posted 12. March 2003 16:07
Cre8,
Thanks! Anyone know the reference to this paper?
IP: Logged
|
|
Cornelius G. Hunter
Member
Member # 81
|
posted 12. March 2003 16:21
Mike:
Here's the citation
Peter Schuster
A testable genotype-phenotype map: Modeling evolution of RNA molecules
In: Michael Lässig and Angelo Valleriani(eds).
Biological Evolution and Statistical Physics. Springer-Verlag , Berlin, pages 56-83, 2002.
See:
http://www.itc.univie.ac.at/Publications/
for all the publications from Schuster's group.
--Cornelius
IP: Logged
|
|
Frances
Member
Member # 169
|
posted 13. March 2003 00:16
In this posting I will show that the data suggest that not only RNA forms such scale free networks but also proteins and that like for RNA the protein networks have the following important characteristics
The neutral networks of any two native folds approach each other to within a few point mutations.
"Neutral networks in protein space" by Babajide, Hofacker, Sippl and Stadler
"Exploring protein sequence space using knowledge potentials" by Babajide et al
Protein have common shapes which are distributed homogeneously forming extended networks that spand the entire sequence space. These results were obtained for various restricted amino acid alphabets including ADLG (ala, asp, leu, gly) which has been proposed as a candidate for a primordial set of amino acids.
Certainly this seems to be not unexpected given that glycine and alanine are found in the Miller Urey experiments.
In fitness landscapes arising from sequence structure maps of biopolymers Peter Stradler argues that
quote:
Neutrality quantitatively changes the dynamics of evolution. While rugged landscaapes without neutral neighbors lead to localized populations, trapping in local optima, and the existence of critical replication rate beyond which sequence information is lost, we find diffusion in sequence space and ever lasting innovation of noverl mutants on landscapes arising from RNA or protein folding.
[ 13. March 2003, 00:24: Message edited by: Frances ]
IP: Logged
|
|
Moderator
Administrator
Member # 1
|
posted 19. March 2003 14:35
I'd like to see discussion on this topic.
IP: Logged
|
|
Janitor@MIT
Member
Member # 125
|
posted 20. March 2003 14:13
Walter Fontana & Peter Schuster, “Chance & Necessity in Evolution: Lessons from RNA,”
http://arxiv.org/PS_cache/physics/pdf/9811/9811037.pdf
[I found especially interesting section 4 on the “reformulation” of Eigen’s quasi-species model.]
Barbel M.R. Stadler, et al, “The topology of the possible: Formal spaces underlying patterns of evolutionary change,”
http://www.tbi.univie.ac.at/papers/Abstracts/00-12-070.pdf
[This is sometimes difficult, at least it was for me, but “The current implementation of the Neo-Darwinian model of evolution typically assumes that the set of possible phenotypes is organized into a highly symmetric and regular space equipped with a
notion of distance, for example, a Euclidean vector space. Recent computational work on a biophysical genotype-phenotype model based on the folding of RNA sequences into secondary structures suggests a rather different picture.”]
Also, Frances, although sometimes this group of researchers states their confidence that the results may be extrapolated RNA to protein, there is reason to believe that the “shape-space covering” principle will be more difficult (and also possibly “neutral confinement” will be more restrictive). Also a caveat about modeling with restricted alphabets and short or compact sequences. I recall a report by Buchler & Goldstein about “designability” with different size alphabets. Sorry no link. And also a report by Erik Bjornberg-Borg (or is it Borg-Bjornberg?) on protein folding landscapes that deals specifically with confinement. Sorry, I’m annoyed with a balky Internet—no link. (I’ll have to switch between machines, but I can provide titles, if not links, if anyone is really interested.)
Also, just from what little I’ve read, some of this related research on proteins seems to relate directly (or indirectly) to Mike Gene’s idea of “hydrophobic transition bias.” (Is that your idea, Mike Gene, or did I just make that up?)
Very generally this body of research raises a sort of “channel-coding” question in my mind: Does it make any sense to say that evaluations wrt “fitness” are independent of any representation or coding? Maybe a sort of principle of “minimal frustration” holds? I.e., that by a “particularly clever” encoding to the (evolutionary) channel we can effectively minimize (or parametrize or bound or however you want to look at it) problems in constrained or frustrated optimization. Make sense?
[I've deleted an absurdly long link to citeseer: Vassilev & Miller, "The Advantages of Landscape Neutrality in Digital Circuit Evolution"--RDP]
[ 20. March 2003, 14:17: Message edited by: Janitor@MIT ]
IP: Logged
|
|
Janitor@MIT
Member
Member # 125
|
posted 26. March 2003 14:36
Codes are algorithms. If we imagine the problem as one of translating data (raw information) into some format for convenient storage, retrieval, and efficient transmission, then a code is an algorithm that solves this problem. Obviously, codes solve other related problems, e.g., they must be reliable or optimal wrt to errors that occur in processing. (Therefore the considerable attention given to “self-correcting” codes.)
But imagine that our problem is not one of defining and measuring error wrt to some absolute fidelity criterion. What if our problem might be characterized as one of an effective trade-off between some level of fidelity and some level of required changes (not really “errors”)? Our “error” would then be defined not in terms of departures from absolute fidelity, but departures from some other criterion of optimality. (Hint, hint.)
Generally we do not consider that a code-algorithm is a solution to more than the narrowly defined problem as stated at the top—reliable information processing. What if I suggested that by a code scheme alone we could potentially solve difficult problems in the local (and global?!) optimization of complex systems? Balderdash! Poppycock! Can’t be done! Yeh, well, and they told Orville and Wilbur that bicycle would never fly.
Can it be done? I have to admit that its not my idea and in fact its already been done!
Someone once said, a cosmologist I believe, that scientists’ problem is not taking their theories too seriously. But not taking them seriously enough. The analogies employed by scientists are also theories. When biologists arrived at the “code” analogy, their problem was not taking it too seriously. It was not taking it seriously enough.
Nothing less than utterly, truly, and seriously brilliant! Wish I’d thought of it!
IP: Logged
|
|
|
Printer-friendly view of this topic