You, by which I mean all the mammals in the audience, are running on an operating system a couple of billion years old.
Your genome is full of primordial libraries, monkey patches, self-modifying code, viral hacks and even containers running a different operating system.
This talk is all about the freakish parallels between cell biology and computer architecture.
My name’s Nick Moore, I’m a consultant, working in software development and systems architecture. I’ve been working with computers for a very long time and Python for a fair proportion of that.
Recently I’ve been working in bioinformatics. Bioinformatics is the study of biological systems using numerical analysis. Biological systems are pretty complex, so typically this analysis requires computers. I’m been lucky enough to get involved in that part.
Thanks to my colleagues at Walter and Eliza Hall Institute of Medical Research and at the University of Washington Genome Sciences for their patience while I’ve found my feet in this field.
Mistakes and oversimplifications are all mine.
The one true rule of biology is that all rules have exceptions, and those exceptions have exceptions, and it’s exceptions all the way down. So some of this talk will oversimplify complex issues and processes but hopefully it’ll be enough to give you a general idea and some inspiration to find out more.
It’s not entirely surprising that cellular biology and computing are a bit intertwined given their overlapping history of discovery:
We’ve had written language, expressed as a sequence of symbols, for several thousand years.
But as our understanding of inheritance developed it became apparent that genetic data was also structured as a sequence of smaller units.
Slowly we came to realize that the kind of linear symbolic information processing we were developing to solve
We’ve learned a lot about cellular biology while we’ve been working out computing, but evolution has billions of years of head-start on us.
Genetic sequencing has taken off exponentially. In 50 years, we went from the discovery of the structure of DNA to having reference sequence of the entire human genome which you can download for free. Thankfully, computing has kept up. It’s hard to imagine how scientists of the 70s would deal with gigabytes of data, whereas these days that’s just a movie.
The personal computing era got started when enthusiasts mailed each other printouts, punch tapes, cassettes, floppy discs, whatever it took to share their discoveries and advances.
Early genetics enthusiasts discovered that Drosophila melanogaster, a kind of fruit fly, could be caught easily just about anywhere in the world, selectively bred and exchanged with other experimenters. A lot of early advances in genetics came from the exchange of fruit flies!
picture of a cell
The basic unit of life is the cell.
Primitive organisms like bacteria are unicellular, each organism is one cell and each cell is one organism, and they reproduce by copying their innards and splitting in two.
A single cell is a bit like a tiny computer.
Inside there’s a bunch of programs called genes which tell the cell how to make proteins, which are the working machinery of the cell.
There’s an outer cell membrane which keeps the insides in and the outsides out. As we’re aware, just running arbitrary programs can get you in trouble. The cell interacts with the world through channels in the cell membrane, which are kind of like ports which selectively let specific molecules in and out of the cell.
Picture of DNA
Genes, the programs of the cell, are encoded as chromosomes, very long DNA molecules not unlike a tape. DNA is built up out of four “bases”, Adenine, Cytosine, Guanine and Thymine, generally just abbreviated as A C G and T.
| base | complement |
|---|---|
| A | T |
| C | G |
| G | C |
| T | A |
They zip together in pairs, A complements T and C complements G, so we generally refer to this smallest piece of genetic information as a “base pair”. This is the unit we’ll use to compare genome sizes. Because there’s four possible pairs each “base pair” is equivalent to 2 binary bits of information. So “1 million base pairs” that means 2 megabits of information.
Bacteria have relatively small genomes, typically a single circular chromosome of hundreds of thousands through to a million or so base pairs. The human genome by contrast has about 3 billion base pairs, and each of us have two copies spread over multiple chromosomes. We’re a lot more complicated than a bacteria, but there’s a species of lungfish with 130 billion base pairs and an amoeba with 670 billion base pairs. So who’s counting?
Unicellular organisms reproduce by copying, and errors arise during the copying process, sometimes leading to novel features in the cells. Good features thrive, bad features dwindle, and that’s natural selection, leading to evolution.
Copying chromosomes can make simple mutations, kinda like typos, and these can also cause a gene to be shortened or lengthened or split into pieces or fused with another gene. Bigger errors can also result in multiple copies of a gene appearing, and then these multiple copies can evolve in different directions. Changes can lead to parts of the genome effectively being commented out.
On top of this, pesky things like retroviruses can introduce new genes entirely, effectively writing themselves into a cell’s genome to get the cell to make more retroviruses.
It’s pretty rare that any of these changes are helpful. Most often, they just result in a broken gene. But over evolutionary timescales, the helpful changes add up.
slide: plasmids
Bacteria also exchange genes with different bacteria or even other organisms, by passing plasmids, which are like mini chromosomes. This is called horizontal gene transfer.
It may seem counter-intuitive to give away your code to your competitors, but consider: the thing evolution is optimizing here is not the organism, but the gene. If a gene which confers, say, antibiotic resistance gets copied into a new organism and that organism thrives, then the gene is successful even if the old organism is out-competed by the new. The gene will get passed on to more organisms.
Yep, Free Software is billions of years old, and there’s been a whole lot of cutting and pasting from the primordial script archive.
central dogma slide
DNA --transcription--> RNA --translation--> Protein
The “source code” of the genome is DNA, but DNA is just a stable storage for the genome, it doesn’t actually do anything. Instead, DNA is transcribed into RNA, and then RNA is translated into proteins. This is often called the “central dogma” of cellular biology.
add some arrows for replication, retrotranscription, ncRNA
There’s also cell replication to consider. Retroviruses can translate RNA back into DNA so we’d better include that There’s also non-coding RNA which functions directly rather than being translated into a protein first.
add RNAP arrow back from proteins to DNA->RNA edge
Transcription from DNA to RNA is done by a protein complex called RNA polymerase (RNAP). This little machine unzips the DNA and assembles an RNA molecule, but RNAP is itself made of proteins.
Proteins are produced from genes, so to build the transcription mechanism we first have to run the transcription mechanism …
add arrows for splicing, ribosomes, tRNA
There’s also splicing and translation to consider. DNA is transcribed into RNA but that RNA isn’t in it’s final form, first it needs to be spliced. This is done by, yep, more ncRNA and more proteins. So RNA is self-modifying code, and one piece of RNA can affect the way another piece is expressed, the proteins it produces.
Before we can compile the compiler, we need to compile the compiler.
In computing, we call this bootstrapping.
You start off by using very primitive tools, possibly even a pencil, to create a very simple first compiler, and then using that compiler you can build a more sophisticated compiler, and so on.
All of this only works because the parent cell contained enough of these mechanisms to get the whole process started. Kind of a boot disk. Those molecules can then make more of themselves, and the cell continues to run.
You can think of these genes, as an “operating system” which the rest of a cell’s biology is implemented on top of.
There’s no documentation or source control, but by looking at the common features of all cellular life we can hypothesize a “univeral common ancestor” which arose about 4 billion years ago, and from which all life is descended.
standard codon table
Translation from RNA to Protein isn’t simple either. A protein is a long chain of amino acids, which is built up by a complex molecular machine called a ribosome, built from ncRNA and proteins.
The mapping from RNA to protein is done by transfer RNA (tRNA). Groups of three bases called codons correspond to different tRNAs which each bring an amino acid molecule to add onto the protein.
This table shows the “standard code” which maps codons to amino acids. Not all organisms have the exact same code: this table isn’t a law of nature. It’s a result of what tRNA happen to be around, and tRNA is produced from the genome, so this translation is happening “in software”.
It somewhat resembles the instruction decoding tables used in microprocessors — there’s some redundancy where 64 codons translate to 20 amino acids. Translation starts at a “start” codon and finishes when it reaches a “stop” codon.
But there’s no looping or branching or I/O or whatever, so how is this like a program?
regulatory image
Well, genes are not expressed equally. When genes are packed into a chromosome, that’s kind of like a library. There’s some header information before and after each gene, known as the “Regulatory Sequences”. These affect the way RNA Polymerase attaches to DNA, and how Ribosomes attach to RNA. Genes, or groups of genes, can be promoted or suppressed. Splicing can be suppressed or altered to produce different proteins.
All this is under the control of the molecules within the cell, which themselves exist through the action of other genes or from external stimuli.
If each gene is a statement, the cell’s “program” is found in the interaction between those statements. And those interactions are extremely complicated.
At some point about 3 billion years ago, a particularly entrepreneurial branch of the Archaea developed a more sophisticated internal cell structure. These are the Eukaryotes.
Animals (including humans) are Eukaryotes. Also plants, fungi, algae and slime molds.
The Eukaryotic genome is protected inside a nucleus, and specialist organelles perform specific functions within the cell. They’re linked together by the cytoskeleton, a network of filaments and tubes which spans the inside of the cell.
You can think of these as peripheral controllers or coprocessors supporting the main processor, linked by buses. DNA to RNA transcription occurs in the nucleus, and then RNA to protein translation occurs in the endoplasmic reticulum surrounding the nucleus.
Yep, we’ve just added a separation between kernel and userland programs.
Also, some of these organelles, the mitochondria, appear to once have been independent Proteobacteria, which were engulfed by the proto-Eukaryotes and instead of being destroyed they were put to work.
Mitochondria are like little containers with their own separate genome. They have their own DNA outside the nuclear DNA. They have their own replication, transcription and translation mechanisms. The cell benefits from their ability to produce ATP, and the mitochondria benefit from the cell’s protection.
But they’re effectively running their operating system, like little containers within the host cell. They do their own RNA translation, using their own tRNA, and they happen to use a slightly different translation code to the host cell. They’ve been engulfed but they’re still separate and running their own OS.
We like to think of vertebrates as the pinnacle of evolution but plants have done this twice! In addition to harnessing proteobacteria as mitochondria, plants have harnessed photosynthetic cyanobacteria as chloroplasts.
You’re probably familiar with the phrase “Embrace, Extend, Extinguish”[^eee] referring to the way proprietary systems can use and then destroy open ones. This may be happening here too: there’s evidence that the functions of mitochondria are slowly migrating into the nuclear DNA and being lost from mitochondria.
[^eee] `https://en.wikipedia.org/wiki/Embrace,_extend,_and_extinguish
Most (not all) Eukaryotes are multicellular, an organism is made up of many, many cells. Every cell in an organism has the same DNA, the same programming, and what role it ends up performing depends on how it is specialized.
You can think of a the DNA as a container image which gets specialized at runtime. The cells in your kidneys contain all the instructions needed to be a brain cell, they just don’t use the brain-specific ones, and vice-versa.
The cells in a multicellular organism each have their own separate existence but also communicate by exchanging molecules. The multicellular organism is a network!
Even weirder, some organisms like slime moulds and some bacteria can form multicellular colonies which act somewhat like a multicellular organism but which can also break apart into unicellular life, depending on conditions.
Okay, great, so biology is hard. Well, mostly squishy. Let’s have a look at a couple more weird overlaps.
There are a lot of things we don’t know about genes, and while we can make some progress doing one experiment at a time, we can make a lot more progress doing thousands of experiments at once.
For example, some of my colleagues having been looking at G6PD.
This is a human antioxidant gene.
If your G6PD isn’t working well, you might end up with haemolytic anemia which is not nice.
We want to find out what parts of G6PD do what, so a good first step is to find out which bits stop it from working if they’re changed.
variants table
G6PD is 1548 bp long, so we can make many versions, each of which has one base flipped to another. That’ll end up with about 5000 variants.
Some of these variants we can predict will work fine because the changed codon still maps to the same amino acid — we call these “synonymous variants”.
Some of these variants we can predict won’t work at all because the changed codon maps to a “STOP” — we call these “nonsense variants”.
Now a gene is kind of it’s own little function so just like in software the first thing is to simplify the test setup. Testing on humans is difficult. They won’t sit still while you bleach them. So instead of testing on humans we can take our variants and tranplant them into yeasts.
yeast picture
Yep, brewer’s yeast, beloved of many a PyConAU speaker. Biologists are very fond of yeasts because under ideal growth conditions, they can reproduce asexually, doubling in population every 90 minutes or so.
Also, you can design an experiment which kills a billion of them and no-one cares.
First we knock out the yeast’s own antioxidant gene, then introduce our variants of human G6PD via plasmid transfection.
When these yeasts reproduce, they’ll keep the variant genes we gave them.
experimental setup
We put them in an experimental setup called a turbidostat which is basically a home brewing setup but designed to keep the yeasts growing forever.
They reproduce happily, and the distribution of variants within the population doesn’t change until … we introduce some bleach.
Yeasts with an effective G6PD variant can protect themselves against the bleach, and continue to thrive.
Yeasts with an ineffective G6PD can’t, and are swiftly out-competed by those that can.
population graph
If we sample the population at intervals, we can see which variants thrive and which dwindle away. And we can give them a score, from fine to terrible.
distribution chart
We can asses our scoring results by comparing the distribution of all variants to the distribution of synonymous variants, the ones we thought were going to be fine, and the distribution of nonsense variants, the ones we were pretty sure were going to be terrible.
variant map
Then we can make a map of these variants, and this map illustrates which parts of the protein are most critical and therefore which gene changes matter the most.
Information like this can give us some idea of the clinical significance of otherwise unknown variants, and help doctors give advice to patients with those variants.
print() slide
Debuggers are nice, but who here has ever resorted to print() (or printf() or console.log()` or &c) just to find out if a bit of code is even running?
LED slide
The embedded software equivalent is the LED.
Things go wrong quickly in embedded code, and your CPU can halt and restart faster than the first H in Hello, World! can make it out the serial port at RS-232 speeds.
So if you want to know where your code is getting to before it crashes, why not light a series of LEDs, one at a time?
jellyfish slide
Debugging isn’t easy in biological systems either. Even RS-232 is actually fairly modern by comparison. Thankfully we have an unlikely ally in the humble jellyfish.
Jellyfish have a gene GFP which codes for a fluorescent protein.
This gene can be located near or fused with a gene of interest and then when that gene is expressed, so is the fluorescent protein.
So cells which express that gene, glow!
emissions table
Since the discovery of GFP, various other genes have been discovered or invented which emit light at different wavelengths, so now we have some serious debugging potential.
dr dinosaur comic
Incidentally, every few months this sort of experiment gets picked up as a whacky “scientists have made a chicken which glows in the dark” article in the papers. A glowing chicken would be neat, but that’s not the whole story.
vampseq slide
For example, a technique called VAMP-seq uses fluorescent markers to sort cells and assign scores by gene expression rather than population.
Messenger RNA, or mRNA has been in the news a lot recently, for all the wrong reasons. A very large proportion of us here today will have benefited from mRNA vaccine technology and a lot of us will get to benefit further as research continues into mRNA therapies for infectious diseases and cancers.
These therapies work by directing your immune system. Your immune system is very effective, but also kind of slow to anger. If you can give it a hint about what things to look for it can respond much more rapidly and effectively.
That’s called vaccination. Historically it’s been done using similar but less deadly infections, such as cowpox inoculations for smallpox, or by introducing an attenuated, inactivated or dead organism for our immune system to learn from without risk of infection.
Ideally we’d like to be able to target vaccination more accurately and develop new vaccines more effectively. We can identify proteins on the outside of cells we’d like to target, and teach the immune system to attack. This process can target viruses, bacteria or cancerous cells.
But those proteins are often not stable enough to use directly. Instead, we can get human cells to make the proteins in place. We can do that by introducing mRNA which codes for the protein we want to target.
But Cells don’t just read any old mRNA they find laying around. Instead, stray mRNA will induce an immune response and get cleaned up by your immune system before all this fun stuff happens — get caught your body’s spam filter, in other words.
Instead, mRNA has to get smuggled into the cell with a bit of a bait-and-switch … just like spam:
Ηο𝚝 Ꮮսхս𝚛у ꓪа𝚝сℎ℮ꜱ
ıո γοᴜⲅ ⍺𝖗ҽа
None of those letters are what they seem, instead they’re a smattering of letters from Armenian, Cyrillic, Greek, Turkish and other alphabets and some other symbols and smallcaps thrown in [^1][^2]. This can help sneak messages past your spam filters while still remaining human readable
[^1] thanks to https://util.unicode.org/UnicodeJsps/confusables.jsp
[^2] this isn’t just a problem with humans either, see https://embracethered.com/blog/posts/2024/hiding-and-finding-text-with-unicode-tags/
Similarly, substituting naturally occurring alternate “characters” such as pseudouridine for uridine in RNA can prevent an immune response to the mRNA, allowing it to be taken up by cells.
| nucleoside | modified nucleoside |
|---|---|
| uridine (U) | pseudouridine (Ψ) 5-methyluridine (m5U) 2-thiouridine (s2U) |
| adenosine (A) | N6-methyladenosine (m6A) |
| cytidine (C) | 5-methylcytidine (m5C) |
Like the unicode characters, the modified nucleosides are different enough to evade spam filtering but similar enough to be “read” by RNA polymerases and used to make proteins, and those proteins go on to produce the desired immune response.
The technique of using pseudouridine to smuggle mRNA vaccines into cells[^3] was pioneered by Katalin Karikó in 2005, used in the Pfizer and Moderna COVID vaccines in 2020 and won Karikó a Nobel prize in 2023.
[^3] https://pmc.ncbi.nlm.nih.gov/articles/PMC2775451/