DNA seen through the eyes of a coder
ds9a.nl
ds9a.nl
http://www.biostat.wisc.edu/~craven/hunter.pdf
I think it's the first chapter of this book:
http://mitpress.mit.edu/books/processes-life
I once considered going into bioinformatics, and did an intense three weeks sprint trying to learn some molecular biology, ending in a seminar presentation to other people explaining the basics. I used this book back then which I also recommend strongly to those interested:
http://www.amazon.com/Bioinformatics-Molecular-Evolution-Pau...
It covers all the basics of molecular biology very understandably and at the same time the scientific/computational content is interesting even for a computer scientist. Still, learning this stuff takes hard work, you have to rehash some relevant chemistry first or you get nowhere, than biologists use a lot of both chemical and biological lingo which you have to understand, and only then the actual biological content becomes clear. Once you do understand it, however, it's beautiful, beautiful stuff, one of the most beautiful things one can learn in general I think, of which you unfortunately won't get a sense from reading this article, or in general from trying to understand it by sloppy metaphors. Do yourself a favour and try to understand this for real.
That's the kind of attitude that pushes competent programmers away from bioinformatics. End result? Almost all bioinformatics software is a steaming pile of crap.
So it's pretty damn good code, then? (Having worked with plenty of coders from the former soviet bloc: they have a pretty low tolerance of crap code)
https://news.ycombinator.com/item?id=6896779
And then there's RNA, which can fold and form structures, sort of like proteins.
Anybody who has ever seriously dealt with things produced by evolution knows that evolution "cheats" and optimizes when constrained by resources, producing structure upon structure upon structure, crazy indirection and non-obvious side-effects, as long as it helps the goal (survival and reproduction usually).
That's why I don't think human brains will be understood any time soon. We might figure out all the layers, but to figure out all the hidden structures would take a superhuman AI.
EDIT:
Example of circuit design by evolutionary algorithms:
http://hforsten.com/evolutionary-algorithms-and-analog-elect...
So, I don't want to rain on anyone's parade, but we didn't "just" find this out. We've known for quite a long time about secondary (and even tertiary!) "codes" in DNA. That HN article was the result of a press-release about something that was interesting, but certainly not earth-shattering new theory. The reason it was in Science was because they did an extremely large-scale test of a whole bunch of different codons on thousands of different gene promoters, and directly quantified the impact of rare codons. That was an impressive way to settle an open debate.
Anyway, it's good to know that there are higher-order interactions encoded in DNA than just DNA -> RNA -> Protein, but you should realize that this is an old/deep area of research. This is pretty much what bioinformatics is about, actually: deducing the higher-order structures in DNA, RNA and proteins.
What? As someone who can write code in several different languages and who has also made de novo designer modifications to proteins at the low level (biological equivalent to assembler, if you will: http://ityonemo.github.io/acs2013/#/, if you're curious) I can attest to the metaphor between the two. It's not totally simple and not one-to-one, but in many situations it has helped me. Hopefully it keeps helping me as I move into my next stint (http://indysci.org/projectmarilyn)
If your idea of "biology" is "going into bioinformatics", well then no wonder your perception of the metaphor is skewed. Bioinformatics is more like "big data on biology"; what is more analogous to computer programming is Synthetic Biology.
Now, I would happen to disagree with many of the metaphors in the OP (and have, as a practitioner of both, better ones in mind that I use in my work), but that there are no connections is, I think, misguided.
One can understand this better by looking for example at the difficulties encountered by people who do actually base computations on the DNA:
It's not clear to me what your link about the DNA code construction really means.
Consider instead:
1) people can encode terabytes of digital information in DNA and retrieve it using molecular techniques. To do so accurately enough requires coding.
2) Crick, who helped figure out many of the basic information coding mechanisms, widely acknowledged being influenced by the theory of information as proposed by Shannon.
3) both computer scientists and biologists frequently use the theory of control mechanisms (feedback, etc, also known as 'cybernetics' when I studied under David Huffman) to understand how their systems work, and engineer new ones.
4) you can make a computer out of DNA< and you can put computers into biological systems; they interact according to the laws of information theory, entropy, thermodynamics, etc.
The relationship between computing theory- as practiced today- and what biology has managed to spontaneously developed- are intimately related at the most basic levels of physics. Understanding that is KEY to being a productive biologist or computer scientist, and escpeially so for bioinformaticians.
I'm not defending the original poster's use of analogy between introns and comments. Those are naive, simple analogies.
What I am responding to is your categorical statement that there are no legitimate analogies between CS and biology, and that's patently false.
I'm certainly not claiming - without evidence - that biological systems perform computation. I strongly suspect that biological systems carry out computation- quorum sensing being a canonical example- and in some sense, any sufficiently complex biological system can be considered an analog computer of some sort, with goal-seeking computational behavior.
One of my interests in some time has been building a digital circuit in DNA; one that can compute, using state, a taylor series approximation of an interesting constant, such as pi, using feedback circuits, error correction, and other "digital" approaches. None of this is impossible, it's just technology. People have already started doing this. That the underlying systems are capable of being turned into computing systems is just an outcome of the fact that biological systems are sufficiently complex that they can be used to instantiate simpler digital computing systems.
On the other hand, I think there are lots of CS people who apply blindly oversimplistic models to complex biological situations without understanding the biology, and that's what I was trying to protest against.
You can't conceive nature as an engineer and talk about how sloppy and random its designs are. You have to bear in mind that nature has made things that we can scarcely comprehend, let alone engineer, so if you're critiquing its designs as if you can read its mind and see what it was trying to do, you're probably missing the point. It pulls off incredible things, so nature is the master and we're the student.
Why not? We can talk about how sloppy something is from a given perspective; we do it all the time, and then try to improve things (there will be no progress otherwise), and if we succeed/fail, learn more ...
>>as if you can read its mind
Nature/evolution is not a sentient being, thus has no mind. Do you have evidence to the contrary ?
http://www.biostat.wisc.edu/~craven/hunter.pdf
is actually the first chapter of "Artificial Intelligence and Molecular Biology":
http://mitpress.mit.edu/books/artificial-intelligence-and-mo...
The "computational" models that have been successfully used to model the DNA have nothing to do with treating it as a description of a computation, they are only computational in the sense that they use mathematics and computing for making predictions about biological phenomena. Those are mostly probabilistic models that treat the DNA as a string of discrete symbols, and encode the physical and chemical complexities associated with this approach in various ways, one good example is Hidden Markov Models, where the probabilities might encode physical or chemical knowledge:
This is one of those instances where the press/pop media can be a bit behind. Some bchem textbooks from even a few years ago are obsolete. Research into DNA-DNA interaction really has become hotter in only the last few years as we've begun pinning down protein roles. There is a whole layer of interaction between epigenetics, differential RNA splicing, and DNA-DNA feedback that is just mind-boggling.
Of course the extent of non-coding DNA that plays some important role has not been known long, and we surely don't understand it yet completely. But finding that some non-coding DNA has a regulatory function has not been a surprise for decades now.
can you provide some evidence there is any non-functional DNA in the genome? Try hard, now, you're talking to somebody who's studied this for 20+ years.
You are already aware, of course, that 90% of the human genome is unconserved, that 50% of the genome consists of dead transposons and viruses, and that mice have been generated that are homozygous for megabase-scale deletions with no discernible effects? Perhaps you can come up with some hypotheses that could explain these facts that are consistent with your claim of functionality?
You've heard of pseudogenes, right? Why do they look exactly like broken genes? What do they do?
Why is there great variability in the sequences of repetitive DNA in many species, including humans? Why do some individuals have many more copies of these tandem repeats than others? Are the extra repeats functional? If so, why does the number of copies change rapidly and stochastically from generation to generation? Do you know anyone who thinks satellite DNA is functional?
The links below will help you get started on the basics of genome biology. Good luck.
http://www.genomicron.evolverzone.com/2007/04/word-about-jun...
http://sandwalk.blogspot.com/2008/02/theme-genomes-junk-dna....
http://www.ncbi.nlm.nih.gov/pubmed/15496924
"Some of the deleted sequences might encode for functions unidentified in our screen; nonetheless, these studies further support the existence of potentially 'disposable DNA' in the genomes of mammals."
Note their qualifications. I'm asking for proof. Making a viable mouse that has a lot of deletions is in no way evidence that the regions are nonfunctional. For example, maybe they deleted a conserved element which has cold-shock response potential, but the mouse was raised in a room temperature environment.
Many of those "worthless transposons" actually seem to play a role in evolution http://www.sciencedirect.com/science/article/pii/S1369526612... and others show that even "dead" transposon probably play a role in the evolution of tail exons: http://genomebiology.com/content/11/6/r59
basically, I know where you're coming from. I used to even believe the dogma. Then I spent some deep time looking at genomes, functional evolution, and biophysics, and came to the conclusion that papers like this: http://www.nature.com/nature/journal/v284/n5757/abs/284604a0... which influenced an entire generation of scientists, are just wrong. Meaningless speculation in the absence of data! Those of us who have spent a lot of time digging into ENCODE and trying to find the really valuable nuggets are starting to come to the conclusion that vast "deserts" of the genome are in fact filled with rich regulatory elements and other functional (including as-yet uncharacterized elements) elements such as RNA genes that classic mechanisms of DNA-evolution-constraint measurement, like Jukes Cantor, are unable to process.
I wasn't referring to active or co-opted transposons, which make up less than 1% of total transposon sequence in the genome. I'm talking about dead, non-functional transposons. These were indeed active and functional in the distant past, but are long since dead. If you have an hypothesis that such sequences in fact have a function, then by all means let's hear it. Likewise with dead viruses, satellites, etc.
It's interesting you mention 'meaningless speculation in the absence of data', since your friends at ENCODE recently embarrassed themselves in just such a way with their pervasive transcription nonsense, as do those who claim function for repetitive, unconserved regions of the genome without any evidence or even an hypothesis, when we have every reason to believe that these are the remnants of formerly-active elements that were subsequently inactivated.
As I said above, it's clear that a small fraction of non-coding DNA has a regulatory role. We've known about this for decades. Recent discoveries don't account for much; RNA genes make up about 4% of the genome.
Sorry if I've been overly harsh in my replies, but from my experience many bioinformaticians have shown themselves to be ignorant about even basic biology, so I took the precaution of going over a few basics.
Anyway, I don't mind that you're harsh. It's pretty hard for people to unlearn their years of misunderstanding the genome! There is a lot of misinformation. Basically, you're just insisting on this line of reasoning: http://selab.janelia.org/people/eddys/blog/?p=683 "ENCODE says What?"
In regards to the ENCODE project's "embarassment", a couple things happened: 1) the press attributed a number of overly aggressive claims to ENCODE. If you read the papers you'll notice they used very hedged statements. 2) ENCODE actually observed some very interesting things. We took their data, and worked on it some more, and we found that a number of their measured transcription tracks actually did represent functional biology. In particular, I can assure you, based on very solid evidence that your estimate that RNA genes make up only 4% of the genome is vastly lower than reality.
The problem is that decades of work in a narrow field, although it makes you an expert, also dulls an outsider's novel look to the subject and leaves you with numerous explicit and, more dangerously, implicit dogmas/assumptions.
It's useful to have these analogies, and to some extent that really do represent true universals. In particular, in reading the history of Crick, I realized that he was a huge fan of information theory, and it helped guide his thinking about how DNA sequences are interpreted and converetd to protein sequences.
However, it can be dangerous to fall down this path. In particular, biology is hotter, wetter and messier than computing. It requires scientists to have extraordinarily flexible brains; I woudl say after many years, I think the people I met in MIT Biology are smarter than the people in MIT CS- their ability to reason over ambiguous data and come up with predictive conclusions is downright amazing.
If you're a computer person who wants to learn more about this, I have a couple suggestions: 1) buy Molecular Biology of the Cell 2) read the whole goddamn thing, slowing down to understand every concept rather than skimming.
The very definition of intelligence is not 'being smart', but having the ability to select one option out of a set of possible options. That is what we, programmers, do. We don't just throw lines of code randomly. We select specific ones for a specific purpose. That's how we build software, mechanisms of information put in motion by the computer. We put our logic into a decisional mechanism which mimics our decisional ability.
However, life does more than that. Life is more than a mechanism driven by a source code. Consciousness goes beyond rules of decision found in programmable machines. But even if you're a physicalist, the abilities that simple beings have such as recognizing objects, paths, building nests, traveling long distances, using tools, are amazing in their own right.
And yet, these all are strong evidence there is no designer beyond it all. It's mere chance, bits on a string selected by nature.
You take the intelligent-design approach - pattern recognition on designed things. I hear you. But what about the compelling inductive evidence we have that deterministic, if chaotic, natural systems are perfectly capable of establishing a feedback loop and modifying themselves?
IMHO, accepting evolution is all about wrapping one's head around the notion that the result of this modification, when stretched over unimaginably large timescales, is significant change.
http://en.wikipedia.org/wiki/Domesticated_silver_fox
By selecting for behavior rather than physical byproduct, they have induced evolution at a rate not thought possible in vertebrates.
Not to mention: http://en.wikipedia.org/wiki/Evolution_of_influenza
Edit: I think the parent comment was about the difficulty imagining the process of evolution which we assume takes billions of years because we cannot observe its impact directly in ourselves except to interpret the results. Yet, it is both observable and inducible in things around us. Not so hard to imagine.
http://www.nature.com/nature/journal/v461/n7268/full/nature0...
Their generation time is shorter, but the principle is the same.
> ...yet, these all are strong evidence there is no designer beyond it all. It's mere chance, bits on a string selected by nature.
Are you saying that you believe that random processes are the designer?
How is that the most upvoted comment on HN? Because honestly, the only way my mind can make sense of that fact is by postulating the existence of an all powerful being able to bend reality to his will....
But it's not just
> mere chance, bits on a string selected by nature
but an incredibly complex and chaotic system ("the universe") with billions of feedback loops, with agents both affecting the universe and being affected by it and their own actions. It's not surprising that with all the processes occurring (many of them [chaotic](https://en.wikipedia.org/wiki/Chaos_theory), if not truly random) we get emergent behaviours, and then even more emergent behaviours on top of those!
You can define "Life" to be the set of self-sustaining phenomena, and there are remarkably few of those. But then that's what makes it exciting!
Large phenotypes (individual organisms) consist of billions of cells each doing their own thing and interacting with others in very specific ways, and every one of these is in turn a complex system of proteins and molecules doing the same, which are all complex arrangements of atoms, etc.
The complexity of life on earth is owed to exceedingly hospitable conditions and the good luck of not one, but multiple levels of self-sustaining mechanisms bootstrapping themselves. But it still took literally billions of random events for those to occur. Each level works according to the interactions of those below, but they each exist due to exceedingly good luck.
i.e. the idea that life is due to nothing more than chance is a strawman
The balls that make it through the wall, make it because of chance or there's something else /I don't understand/?
For example, we might say that children born with fatal/harmful defects are "unlucky," which is true, but this is a short-term perspective. Over larger timescales, we can say (poetically) that the rarity of these particular kinds of DNA configurations is a "design choice" rendered by the laws of nature. The larger the time scale, the less meaningful individual "chance" things are and the more "systemic" factors come into play. This is probably an instance of "the law of large numbers".
--
Another important thing to point out is that the environment is dynamic and species interact with eachother in all sorts of ways. The concept of a single species evolving in vacuo is very dry, but the reality is different. Throw a few different species into the mix and suddenly you have an explosion in the kinds of behavior and phenomena you will find. I think intelligence is one of those phenomena.
In other words, the natural world has all the depth and richness necessary to account for the "enlightened" properties that you attribute to life. In fact, if I had to pick one gripe I have with religion, it would be that it blinds people as to the raw amazingness and beauty that can be found in the natural world.
Here is a question. Do you agree or disagree that survival/death decides which of nature's experiments work?
I see you haven't met my coworkers.
P.S. I'm self-employed right now.
Natural selection acts on randomness, natural selection itself is the antithesis of random chance. I mean no offense, but your statements reveal a lack of understanding of the potential and strength of evolution by natural selection in general. It is elegant in its simplicity. Basically, given a stable replicator, make variations on it and those that continue to replicate will necessarily have some differential in survival, sometimes due to the variations. This is clearly explained in a segment of The Blind Watchmaker [1], where generations of an aerofoil shape are varied at random, but selected for their lift potential. Please give the short segment of the video a few minutes of your time, I think it is a wonderfully intuitive explanation of the core of natural selection.
[1] http://www.washington.edu/news/2013/12/12/scientists-discove...
(ref: I used to be a bioinformatician)
Secondly, within the world of practicing DNA scientists, the link you provided was 'laughed out of court' as being the stuff of PR dreams, but not actually true. For example, see http://www.bacteriatobonobos.com/1/post/2013/12/go-home-pr-t...
[0] only because so many times I've relied on the bioinformatic analysis/annotations and actually figured out what was going on by looking at it by myself, and then thought, goddammit, if the bioinformatician ever spent time in the wetlab they'd never have missed this! Not saying that a wetlabber could have competently written a script either (they'd either be too dumb to or too lazy to - that's what informaticians are for). Anecdotally, Ham Smith and Clyde Hutchison, used to write their own fortran programs for identifying promoters, ORFs, and what not. A few weeks ago, they realized they had totally screwed up an analysis because the excel sheet of ORFs they had was mapped to the wrong sequence by the bioinformatician they handed it over to...
GRN is.
https://en.wikipedia.org/wiki/Gene_regulatory_network
https://en.wikipedia.org/wiki/Gene_regulatory_network#Modell...
GRN can be modeled using different levels of abstraction and accuracy as boolean network , recurrent neural network or as stochastic gene networks.
In other words, they are capable of complex computations, but computational model looks more like neural network or stochastic network.
I wonder if we turned this back around, would it suggest some novel designs for computer systems?
After years of being a senior dev and such in some web shop, I'm actually starting a job in bioinformatics in a few weeks, it's beyond exciting. It's articles like this that remind me why being a programmer can be interesting beyond the code. We live in very interesting times.
http://en.wikipedia.org/wiki/Logo_(programming_language) http://en.wikipedia.org/wiki/G-code
As a software engineer/developer/programmer, coding is just one aspect of what I do.
Venter's doing it right now.
Given enough time and energy, you can build an organism from scratch. Of course, Venter's still pretty far away from that goal. But I suspect he will succeed; we all know there are none but technical barriers.
My wife and I managed to do it a couple of times. Not that hard.