Hacker News Reader

Comment by dboreham | Hacker News Reader

Sometimes little frustrating that people writing articles in this field seem to have no exposure to computer science.

This is the gene equivalent of a filesystem: the DNA fragments are like disk blocks. The interposing sections are file metadata. The shearing mechanism is the filesystem reconstructing a stream from the lower layer blocks. There's probably some redundancy and error correction in there too.

It needs a filesystem for the same reason they were invented for computers: to provide an impedance match between the upper layer semantics (a stream of pairs describing a protein) and the lower layer storage (blocks). Using block structured storage is more flexible in terms of being able to insert in the middle of a file, etc etc.

View on HN

That's a weird thing to get frustrated about...that someone in a completely unrelated field didn't have the experience or courtesy to explain something using analogous terms to the thing you happen to be an expert on.

Doubly so, when the two domains don't actually map cleanly to eachother. DNA is not a computer program, or a file, or storage. There's no real distinction between data, metadata, and 'code' in it, either in structure, or in practice.

I don't think they are unrelated fields. Computer science is the study of computation. DNA, to me, is clearly a quaternary computation system.

I think there is a lot for both fields to learn by studying knowledge from each. Bioinformatics seems to be on that track.

Cool so what's the last thing you described to your team using terms from DNA research?

genetic optimization algorithms?

We've got DNA which is basically a storage system. RNA can be catalytically active on its own. Typically RNA and proteins, or complexes of such, act on DNA in various manners.

Maybe you're right in some way, but also consider whether using the nomenclature and ideas used to described processes of DNA repair, transcription, and translation in biology to describe electronic computation works well. If it does work well, then maybe the reverse would also work well. If it leaves much to be desired, then consider the possibility that computer science ideas may be too specific to electronic or mechanical computation.

I don't like appeals to authority, but as a computational biologist, I am both a computer scientist and a biological scientist. So understand that I'm responding to you as exactly the kind of person you think should be drawing the kinds of links that you're suggesting.

I'm sorry, but this is just not a reasonable analogy. DNA sequences are not like computer files. The reason that they're distinct modules in these sequences is due to the need for evolution to be feasible. And also for the basic reason that the sequences are linear and so modules tend to appear in these linear sequences. But there also tremendous nonlinearities as well. Things at very long distances can be importantly related to each other.

The introns are not metadata. There are regions that can be removed selectively and in combinations to cause diversity in the produced proteins. That diversity is advantageous because it allows a single DNA sequence to present many different proteins that are typically related, but can be very different structurally. This splicing capability has evolved apparently from entities that can be seen as endogenous viruses or DNA parasites that have the ability to insert and splice themselves out of DNA and RNA sequences. In many confusing words, that's what the article or pointing at is talking about.

The introns do provide a kind of redundancy, but only in the sense that there are areas that can be modified with minimal effect on cellular function, at least relative to modification of the exons, which directly correspond in a one to one way with proteins.

There is error correction. It's called homologous chromosomes. Everyone talks about there being one genome, but in most complex forms of life, you have more than one copy per cell, usually two, and often more. These multiple copies, in addition to allowing for recombination and sexual reproduction, provide templates on which errors which arise during life can be corrected. However, there are no error correcting codes.

If you'd like to learn more about the actual details of these systems, I strongly suggest an undergraduate molecular biology textbook. The best one in existence is called the Molecular Biology of the Cell.

There are indeed many similarities between computing systems and biological systems, but the analogies you are making don't appear to be clear. Read a book like this, deeply and slowly, and it might change your life. At very least, it'll mean that the world you live in is much less mysterious and much more exciting.

"Molecular Biology of the Cell"

Shout out to this book. Not only has the most amazing imagery, but you can learn a lot just by skimming the book, because of how well organized it is.

Aren't introns like pieces of code that have been commented out?

But they also change the 3D conformation of the DNA itself, which changes access by transcription factors, etc.

It's more like someone during the night cut your magnetic tape in the middle of a txt file tape and glued a picture of a cat between the two parts. The picture of the cat has some special code in the extremes, so it automatically disappear when you open the txt file.

There are some weird case, where the same txt file has two cat pictures, and sometimes instead of removing the two cats, the system removes also the texts between the cats.

How's that different than the special code that tells you to ignore the comment?

You look interested in the subject, but I recommend to read a few biology books about it. There are many weird low level features of DNA that are not so cover in popular discussions [1] [2]. But I don't remember any that is similar to a filesystem as you propose. Take a look, you will be gladly surprised.

[1] One of my favorites is that the bases of ADN are translated in groups of 3 to amino acids, so the code reads like

  AAABBBCCCDDDEEEFFF

It's very unusual, but there are some virus that read the same part in two ways, with different offset, so the same part is interpreted as

  -JJJKKKLLLMMMNNN--

I don't remember if they use the other offset too

  --PPPQQQRRRSSSTTT-

[2] Another, not so interesting but relevant. Eukaryote has linear DNA, so they have some special repetition in the extremes. The idea is that the extremes are difficult to copy by the usual enzyme that copy the main part that has assorted code. But the extremes have a special easy pattern, so the cell can use some specialized enzyme to make them longer.

There are organisms with almost no introns, no redundancy, no CRC, no "metadata" and even overlapping genes to save space, like Giardia genome ( https://www.science.org/doi/10.1126/science.1143837 ). Lots of virus have all their genes encoded without introns, and almost all the genome is encoding something.

I've never seen DNA as close to a filesystem, and our current best bet on introns functions are they are used to create alternative splicing products from the same DNA chunk. I cannot identify this function in a filesystem, where you can obtain two or three different valid files from the same data just by skipping some blocks.

I can - It's called COW Snapshotting. Modern filesystems like ZFS and BTRFS don't ever overwrite parts of the file that change. They abstract it away by keeping an ordered list of blocks. Snapshots are simply copies of the old list.

The analogy doesn't go very far, however.

If a filesystem worked like alternative splicing https://en.m.wikipedia.org/wiki/Alternative_splicing , you could for example store 10 different, valid and useful images using the same disk space. The filesystem would decide how to skip a block or two to give them all.

A woodworker could make a better analogy: DNA is like a box of leftover wood. You can chose the "exons" and build a closet with them, a different sort would make a table.

My point is we CS shouldn't feel pity for the biologist not being able to draw analogies with our field. You don't need an analogy when you are already the best at understanding the system directly.

I like to explain ML to biologists as an analogy with neurons, but it only can go so far. As soon as you need to teach the details things are no longer so similar. The reverse is also true: you can explain a neuron to a CS as a ML neuron, but when you need to explain neurotransmisors in detail, the analogy doesn't work anymore.

Okay, ZFS also does that - ZFS Deduplication keeps an index of each block by it's checksum, and when it detects a collision, it re-uses the existing block, reducing the number of copies - https://docs.oracle.com/cd/E36784_01/html/E39134/fsdedup-1.h...

I'm not arguing at all that the analogy is a poor fit. Just that there are several instances of that kind of technique in CS. See also: Compression algorithms that include backpointers, so that medium-sized repeated runs (like common parts of filenames or fileheaders) can be represented by saying "The next 8 bytes are the same as the 8 bytes 32 bytes ago".

The whole point of an analogy is to make it easy to understand some complex idea using another simple concept. I would use instead this analogy: "alternative splicing is like having a box of Legos (exons) which you can choose in different order and quantities to get different objects (proteins)." I find it more clear than "alternative splicing is like ZFS deduplication and collision detection algorithm, where blocks can be reused if they are identical".

Introns are not for space saving or for efficiency. They provide variability. Is like ZFS could give you slightly different files from what you initially save. E.g. you store feral_cat.jpg spanning a number of blocks, and then just skipping blocks you can get back: feral_cat.jpg, persian_cat.jpg, egyptian_cat.jpg, russian_cat.jpg, alien_cat.jpg, tabby_cat.jpg... You never store "persian_cat.jpg" and try to save some genome space reusing blocks. persian_cat.jpg is just a possible useful image that appears magically from the original feral_cat.jpg.

COW just dedupes, it doesn’t produce alternatives. Maybe a filesystem that figures out how your spreadsheet can be stored partway into an executable, with no loss to either? Yeah, this analogy doesn’t seem real helpful.

What is CRC?

Metadata is everywhere... histone modifications, glycosylation, etc..

Cyclic redundancy check? And there is almost always basic redundancy in non-viral organisms (and many viruses) in that DNA is typically double-stranded, and most organisms have repair machinery that can rewrite across single-strand lesions using the opposite strand (this fails at double-strand lesions).

I don’t want to sound dense here but why is it frustrating that people who write about genetics aren’t familiar with computer science?

I can have similar thoughts at times. People in one field have their own lens to see the world and might miss some structures / patterns that exist in other domains. I felt it was a bit pompous to read a few medical books about the cardiovascular system, a lot of ceremony to describe an organic pump. You'd have to read mathematically inclined papers to start reading about equations and principles rather than latin nomenclature. Which I think is what the grandparent was wishing for.

ps: I absolutely do not put computing above other fields though. I just wish for some pragmatic polymathism sometimes.

I suppose because they spent 6-8 years of their early adult lives focusing exclusively on biology and not computer science

I don't think that's a great comparison. Exons already have sequences that mark 'block' boundaries; amino acids are encoded in triplets of base pairs, but sort of like you see in 8b/10b encoding, there are sequences that are valid but only used for control purposes and don't correspond to amino acids.

It needs a filesystem for the same reason they were invented for computer

Why would that be true? It is a cool analogy but implying that it is the reason requires much more.

Without taking a stance on whether it's right, I believe that is explained right after your quote cut-off.

That the biology of processing DNA is subject to similar constraints and solutions computer have dealing with disk storage is remarkable.

Justification is needed.

At a coarse level, both are affected by information theory, so some parts may look vaguely similar. Sure, it’s plausible you’d find related solutions, especially if you really squint.

But it’s like trying to explain how atoms work using a solar system analogy. It might help with the really easy stuff, maybe? (orbitals) But sticking with it makes going any deeper pretty confusing.

> It needs a filesystem for the same reason they were invented for computers: to provide an impedance match between the upper layer semantics (a stream of pairs describing a protein) and the lower layer storage (blocks). Using block structured storage is more flexible in terms of being able to insert in the middle of a file, etc etc.

As a non-CS person I find this explanation opaque.

Nope, at best the filesystem is an analogy. Just like the stretchy rubber mat isn't a perfect description of spacetime.

https://en.wikipedia.org/wiki/Curse_of_knowledge

https://xkcd.com/793/

Reply on news.ycombinator.com