No individual human's genome has ever been completely sequenced. In addition, there are some parts of human DNA that have never been unambiguously integrated into a larger map of humans' DNA.
The was flag killed. I vouched for this because I had the same question. What does it mean to have sequenced the human genome? If different individuals have a different genome, what is the human genome?
You're right, and the next frontier in genome assembly and mapping algorithms is using graph-based genomes, where differences between individuals or populations can represented as different paths through a sequence graph. (There are lots of better explanations with pretty pictures out there with a little searching)
That's more of a representation problem, isn't it? I'm not seeing where it would require any new development to produce such a graph - just a lot of CPU time to assemble one from a collection of FASTA files, or whatever linear representation. (And a lot of wall-clock time, to convince people to use it...)
Right - It's not so much algorithmic development as development of the toolchains that will enable such representations to be actually useful for downstream analyses.
The article acknowledges this, though perhaps obliquely:
> “As a matter of truth in advertising, the ‘finished’ sequence isn’t finished,” said Eric Lander, who led the lab at the Whitehead Institute that deciphered more of the genome for the government-funded Human Genome Project than any other. “I always say ‘finished’ is a term of art.”
The phrasing of your parent is arguably dismissive and does little to further substantive discussion. Can one meaningfully discuss "the human genome"? As you point out, each human (with the exception of identical twins) has a unique genome. I think it's reasonable and useful to be able to say "the human genome" as opposed to "the chimpanzee genome" or "the drosophila genome".
If one would rather insist on using the indefinite article, i.e., "a human genome", one can present it without resorting to a rhetorical question replete with scare quotes.
Not only do different individuals have different genomes, each cell from the same person will have different genomes. It is more like an "average" human genome, one that no actual cell has ever contained.
Mutation rates are high enough that we should expect at least a few every division. There are supposedly n~6e9 base pairs[1] and mutation rate is said to be p~1e-8 per bp per division[2].
Assuming each mutation is independent, etc, etc we get a back of the napkin via the binomial distribution with mean = n*p ~ 60 mutations per division.
Thanks, actually the wikipedia link was "per generation" and they assume ~ 100 divisions/mitoses per human generation. So it should be 1e-10 mutations/bp/division.
There are somatic mutations (vs. germ line, which would be inherited).
Somatic mutations are just errors in genome duplication in somatic cell lines - i.e. not germ line. These accumulate over time and, yes, can be oncogenic. Many others simply have no consequence at all.
For many organisms, the germ line is 'sequestered' fairly early in development, in part to limit the number of replications needed, and thus errors that accumulate.
From Wikipedia: "The "genome" of any given individual is unique; mapping the "human genome" involved sequencing a small number of individuals and then assembling these together to get a complete sequence for each chromosome. Therefore, the finished human genome is thus a mosaic, not representing any one individual."
Members of a species share most of their genetic material with each other, but there are known areas of high variability between individuals. So to build a specific individual's sequence, it seems like they take the baseline "human genome" data, and overlay the areas that are known to change.
It's strange that this article ends as an advertisement for PacBio sequencing (which can ~50k-60k base reads) but makes no mention of Oxford Nanopore (which has gotten megabase reads and keeps improving). Single molecule nanopore sequencing is on track to sequence across the centromeres of human chromosomes in the next few years.
I expect us to use the long reads from ONT and PacBio to discover that structural variation is even more important and common than previously appreciated. If that's the case, you'd be silly to use the Illumina technology for anything more complex than population genetics, and this will probably trickle into many parts of the market as the medically-relevant parts of the field realize what the basic science is demonstrating.
This is so cool. I'd been taught in high school that the human genome was no longer a mystery, and that science had fully decoded the entire thing. It seems pretty obvious now that the subtleties would be even more complex than we could have imagined.
The fact is, unfortunately, that Nanopore sequencing (and also, from what I’ve read, PacBio) has a dramatically higher error rate than Illumina sequencing-by-synthesis. In the near future, anyway, I would expect to see inaccurate PacBio/Nanopore long reads being used as scaffolds for accurate Illumina short reads (in fact, this is already happening). Illumina won’t be going anywhere any time soon.
>The fact is, unfortunately, that Nanopore sequencing (and also, from what I’ve read, PacBio) has a dramatically higher error rate than Illumina sequencing-by-synthesis.
This is true, but only for Insertions/Deletions. The substitution error rate is comparable to Hi-Seq.
So it's good for resequencing but without a reference or decent scaffolds, you're in the dark.
I've only sequenced a few samples on a MinION. The dominant error seemed to be miscounting the length of a homopolymer stretch (e.g. CCC->CCCCC) but the substitution error rate was also clearly worse than Illumina. I've gotten used to less than 0.1% error from DNA sequencing on a HiSeq 2500. My guess is that the MinION data had 1-3% substitution errors.
PacBio and ONT are far from upstarts, they have both already invested millions and millions into their respective technologies.
Pacific Biosciences has a Market Cap of 532.88M
Oxford Nanopore is at least as big.
They currently specialize in different things. The larger Illumina instruments have a huge up-front cost (~$1M), tremendous throughput, low cost per base, short reads, <1% base error, and significant latency between starting sample prep and getting results back. Nanopore sequencing has trivial up-front costs, same-day results, orders of magnitude higher cost per base, >10% non-independent error rate and very long reads.
ONT will be much more useful for assembling reference genomes (which can then be polished with short reads), characterizing large scale structural variants, and real time applications like "what pathogen am I infected with?".
The simple explanation is that they are different products that only overlap to a limited degree.
If you're doing a standard RNA-seq experiment, short 100bp to 200bp Illumina reads are just dandy for what you need. Most of the time, at least; you could, for example, be interested in different splice variants for a gene family; in that case longer continuous reads are valuable. But then it's no longer a standard RNA-seq.
PacBio and pore-type sequencing is most valuable for polishing reference genome assemblies. For instance, this is a huge issue for crop plants because they have huge, repetitive genomes and are often polyploid. So there's a lot of interest in creating high quality assemblies not just for e.g. corn, but the major research inbreds (B73, Mo17, etc.) as well as private breeding lines. Some of the most valuable crop QTLs (R genes, microRNAs, transposon fragments, etc.) are just 'junk' sequence that plays a regulatory or defense role, so it's super important.
In general, the Illumina data is very high quality and very inexpensive, and works beautifully provided you have a high-quality assembly to work with, and what you're interested in isn't very repetitive. It has many good applications, so I don't see much risk at this point.
If, at some point, longer reads can be done with high fidelity and with similar cost, then there would be an issue. I don't think that's going to happen anytime soon, but I'm sure Illumina is looking at that long-term.
I'm not sure how many they've actually sold but i do know that they have a loan/test program where you can pay a relatively small amount (around $2000 or less) to get set up with a MinION kit. The main recurring cost is the flow cell. I'm currently trying to figure out how to join together short multiplexed samples (150bp - 200bp) and get it sequenced on the MinION. Current short-read sequencers from Illumina or ThermoFisher cost orders of magnitude more than Oxford Nanopore's offering. Hopefully someone else is trying to use the platform for short reads.
I got mine as part of the beta almost two years ago. It's been in release since, and seems popular in the community. It's not Illumina HiSeq popular, but the cost is low enough that lots of people are experimenting with it.
Nanopore sequencing is definitely "already here" in the sense that they're selling a lot of flow cells and people are doing useful research with ONT instruments but it's also a niche technology compared with Illumina.
Interesting, I haven't tried it on the newest chemistry but last time I did a flowcell (earlier this year) that was the INDEL error rate we achieve.
Substitution error rate was very low, but you should evaluate INDELs specifically, rather than just identity like you did on your git, because it is the major drawback of this technology.
I've personally prep libraries and used nanopore. While there is a lot of promise to the technology, I found it to be too inconsistent. Nanopore claims they can get 10gigs of data, while power users claim they get 2-5gig on average. I consider myself lucky if I get 2gigs of data.
The non-random nature of their error profile is also a problem.
Are those experiences on recent chemistries/flow cells? My first library prep gave me ~4 gigabases on a R7.4 flow cell and that was pretty low compared with everyone else in the workshop.
Also, which library prep are you using? I think the ligation kit has historically gotten better yield (not sure if that's true any more).
I agree about their communication and support. It does seem like they put all their support behind the few big labs in the UK that are actively using nanopore (free R&D for them I guess). I get a sense of "if you can't get this to work, then you are just a bad scientist" vibe from them.
Are you working on cell lines or bacteria? Genomic dna? I find most people that tell me they get good data yield are working on samples where dna extraction is relatively straightforward and dna yield is abundant.
I used the flowcell version previous to the current one and got ~2gb of data. I used the ligation kit on genomic extraction of ~10 individuals (crustacean).
In other words they don't know what they're doing and they're making it up as they go along -- "yeah, that part there is probably unused", "oh, wait, it's important after all".
Well: "“A lot of people in the 1980s and 1990s [when the Human Genome Project was getting started] thought of these regions as nonfunctional,” said Karen Miga, a molecular biologist at the University of California, Santa Cruz. “But that’s no longer the case.” "
And: "“I’m between agnostic and a little skeptical that these bits will be important for disease, but maybe I’m saying that because we can’t read them,” Lander said.".
The "no such thing as junk DNA" claim comes from a paper which is, to say the least, controversial in the field. I would hesitate to take it as prima facie reliable, and in general would recommend looking with extensive skepticism upon popular reporting in this realm of scientific endeavor. After all, this very thread originates in a piece of popular reporting that only seems novel because all of the previous popular reporting has been wrong...
As usual, the journalist spends many paragraphs painting a picture of human conflict before actually getting on to the interesting claim.
> The reason for these gaps is that DNA sequencing machines don’t read genomes like humans read books, from the first word to the last. Instead, they first randomly chop up copies of the 23 pairs of chromosomes, which total some 3 billion “letters,” so the machines aren’t overwhelmed. The resulting chunks contain from 1,000 letters (during the Human Genome Project) to a few hundred (in today’s more advanced sequencing machines). The chunks overlap. Computers match up the overlaps, assembling the chunks into the correct sequence.
> That’s between difficult and impossible to do if the chunks contain lots of repetitive segments, such as TTAATATTAATATTAATA, or TTAATA three times. “The problem is, when you have the same exact words, it’s hard to assemble,” said Lander, just as if jigsaw puzzle pieces show the same exact blue sky.
> In 2004, the genome project reported that there were 341 gaps in the sequence. Most of the gaps — 250 — are in the main part of each chromosome, where genes make the proteins that life runs on. These gaps are tiny. Only a few gaps — 33 at last count — lie in or near each chromosome’s centromere (where the two parts of a chromosome connect) and telomeres (the caps at the end of chromosomes), but these 33 are 10 times as long in total as the 250 gaps.
I recommend reading HN user eggie's comments on this from last week.
https://youtu.be/fCd6B5HRaZ8 is the best visualization of how the most popular type of DNA sequencer works (that I've found).
Imagine you have a string of length 3 billion made by randomly choosing from 4 characters. Like this
dna = ''.join(random.choices('atgc', weights=[30.9, 29.4, 19.9, 19.8], k=3_234_830_000))
you get to randomly sample 1 billion[3, page 7] overlapping substrings of length 200[3, page 7] with .1% of the characters randomly changed[3, page 8]. Trying to find the original string from this is technically an undecidable problem. If there's a sequence 400 characters long that repeats multiple times, how could you know if it repeats 5 times or 50 times? (this would be unlikely to happen with random.choices() but DNA isn't random). This is called sequence alignment and it's one of the hard problems in bioinformatics[4].
Thanks, you’re right. I wasn’t aware of that. The experience is rather crappy though. Cumbersome to scroll on a smartphone screen and I’d love to see to entire statement not parts of only at a time.
Seriously. I can't go a week in HN without reading someone's complaint about not being able to read a quote, probably on mobile. If users keep complaining regularly, it's a problem with the software interface, not the users.
I emailed dang about it a while (years) ago. He said he's added it to the fix list, after initially saying that having a quote tag would mess up the site's "character." You may want to email him too.
I fail to see how a quote tag would drive away the throngs of Silicon Valley elites, people who begin sentences with “I’m not an expert, but...”, hardcore Randians, and Paul Graham sycophants.
And yet here you are, on the site whose users you are attacking, commenting on an off-topic comment thread. Perhaps you should rethink your life choices?
You forgot the folks who turn every article on the latest Google phone into a discussion on Apple products. See the latest posting about the Pixel 2 as an example.
Very nice movie indeed. This Sequencing by Synthesis is indeed the most popular method, the words in this type of DNA sequencing are usually 150 base bairs long, but by using fragments that can be much longer (say 2500 base pairs) and sequencing from both ends (as the movie explains) one can derived information on the position of 2 150 base pair long "words" relative to each other.
The alternative technologies, sometimes called third generation sequencing are nanopore based (PacBio and Oxford Nanopore technologies), they work by pulling long DNA molecules through tiny holes and read the sequence directly, more specifically, they read a couple of base pairs at a time and infer the sequence from the current between the sides of the pores. This is very error prone and the (around 1 in 10?) random errors are corrected by sequencing every part of the genomic area of interest multiple times.
To be clear, most sequencing experiments (in research and in the hospital) are designed to look at variations we can understand and thus the "words" (we call them reads) are aligned against "the" reference genome (as opposed to assembling de novo), which is the genome for as far as we understand it (the last official version is from 2013 but is updated regularly with new parts). Typically, a significant portion of the reads a sequencer produces cannot be mapped to the reference, probably as a results of unmapped regions indeed or perhaps deviations between the patient and the reference genome, this is of course even worse in patients with tumors with unstable genomes. This field still has a lot of challenges to tackle.
Nope, not randomly. Having a random error profile is the best case. In reality these machines suffer from systematic errors, especially the Illumina ones. That means that they always fail to read through certain pieces of DNA, or make the same error every time.
This person also forgot to mention that the original human genome project was not even the sequence of a single individual- it was several people.
The other thing this article doesnt mention is the idea of coverage. Its not like a jigsaw puzzle with one copy of each tile-- every time a whole genome is sequenced it is done so 30 times over to ensure that the pieces are correct.
Finally, Read length varies from technology to technology. This one claims average 10k base pairs.
My biologist friend isn't patient enough to ELI5, but AFAIU don't genes realign through the natural course of replication? That is, there is no specific, canonical sequence at large scales, neither between two individuals nor even two cells, even if the complete set of base pairs[1] is otherwise identical (i.e. no mutations).
I realize that the magnitude of this natural shuffling, even during meiosis, is nowhere near like what happens when sequencing DNA[2], but (assuming I'm correct) it's helpful to realize that this problem of deciphering the "correct" sequence reflects something intrinsic to the operation of the machinery; that this applies even when the relevant macro sequences are all genes that directly code for phenotypes as traditionally understood; and that gene sequencing works precisely because many (most?) genes are relatively tolerant to being spliced in at random locations in the strand. (Which is different than saying all locations--or even any two locations--produce identical results.)
[1] I refrain from saying "complete set of genes" because I imagine for some genes location at both the micro- and macro-scale is everything. AFAIU the term gene is sometimes better understood as more a description of the result of possibly non-local sequences which maintain reproductive affinity than an identification of a specifically adjoining sequence of base pairs.
there is no specific, canonical sequence at large scales, neither between two individuals nor even two cells
This is correct, and is a reason why genome informatics has been moving away from representing the genome as a single linear "reference" sequence of 3 billion nucleotides, and towards a graph-based model: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5411762/.
They do that for the rest of their audience, for whom that is important motivating context, or just informational context. I think It’s a good approach to writing, and it’s he really awkward to their main readership otherwise.
There’s even evidence that this is true, for who knows their audience better than them? You, while making up a a portion of the audience, are not the main portion, or so it seems, unless they don’t understand their audience that well after all... although that seems pretty unlikely
I worked on a slab gell dna sequencer contemporaneous to the humane genome project. It's actually worse than described, because the dna isn't cut randomly. Instead, restriction nucleases cut the dna at certain fixed sequences. Of course you can use multiple restriction nucleases, either in the same (subject to their chemistry being compatible) run or over multiple runs. When everything went well, our tech allowed you to read 800-ish bp with good quality and perhaps up to 1.1k if you were willing to to get dodgy quality at the ends.
So you have two problems. The first is, you never read anything longer than say 1k bp ever. The second is say your nuclease cuts at GAATTC. You will have great difficulty counting runs of that sequence.
I think the entire tech that I worked on has been superseded by better technologies. However, I (think?) the limitations are still of interest because they influenced the original sequencing and probably a lot of the early data.
The other takeaway I had, as someone who came to this with a CS and stats background, is that chemistry sucks. None of it is deterministic, and when people say some chemistry does X, what they really mean is it generally does that. Most of the time. Subject to conditions and a reasonable amount of stochasticity. We're built on a pile of random muck hacked together.
Older tech stays around quite a while! No sequencing technology is perfect, so even Sanger sequencing is still used quite a bit for various tasks from clinical sequencing to validation studies to anywhere you just want to phase something that's longer than 300bp, etc.
And yes, the chemistry that biology takes advantage of means embracing uncertainty in the world. It's all thermodynamic chances, with lots of non-linearity added on top of it all.
One of the things the HGP revealed is that we have way fewer genes than anyone expected. Given that the amount of data encoded in genes appears to be far too small to account for the sophistication and variability in humans, it strongly implied that large areas of DNA that were previously considered marker/filler DNA were of great epigenetic importance.
There are also a lot more steps between 'genomic dna' -> 'protein' than was previously appreciated or thought to be of importance. There are a lot of ways to regulate and produce new protiens and variation that is not itself a genomic change.
Alternative splicing, readthrough regulaton, multiple kinds of ribosomes, post-translational regulation, etc. all produce significant variation outside of the dna itself.
It always seemed insanely stupid to me every time I read something about people claiming certain genes being "non-important" or "non-functional"
We have about a 0.0001% comprehension of this whole system, how can you possibly be so arrogant as to think you KNOW that certain parts of the sequence are not important or used in _any_ way? Then again, I'm definitely not qualified to comment on the topic.
Slightly off-topic, but this semi-animation of DNA replication machine that is working right now in your body (in amounts of trillions of devices) blew my mind few months ago.
> * A gene called ARHGAP11B, which was created by one such duplication, causes the cortex to develop the myriad folds that support complex thought; SRGAP2C, also a duplication, triggers brain development.*
A question I've never seen addressed: is one justification for junk DNA to create space for beneficial mutations?
Obviously some mutations are actively harmful, but the reason most mutations are destructive is that they break something already-useful into something inert. That's not a risk with non-coding (or irrelevantly-coding) DNA, which means that in return for spending energy on copying junk DNA, we get space where even barely-useful changes will be net positive.
Is this obvious to experts? Completely stupid for reasons I don't know? Even a possibility worth discussing?
If I recall correctly from my bioinformatcs classes (years ago) - the non-coding DNA still serve many functions - a protective role by containing sequences which stabilize the structure of the DNA, or providing binding sites used in promoting or suppress gene expression. The non-coding are recognizable by different characteristics from the coding sections. Not to mention the fact that by definition, at least half the sequences are just the complement of a coding sequence - so putting in stop codons before and after on the complement would be useful to make sure it's not expressed. Add in structural sequences to create preferences of where it's safe to swap genes with the sister chromosome DNA during meiosis, and all sorts of other things, and you end up with quite a but of non-coding DNA still serving a purpose.
Perhaps not exactly "junk dna" but you are right in your thinking about gene duplications... duplications are thought to promote "neofunctionalization" (a new function) or "subfunctionalization" (specialization).
>A question I've never seen addressed: is one justification for junk DNA to create space for beneficial mutations?
Yes, gene/genome duplications can help create raw material for adaptation, or be adaptive by their own right.
>That's not a risk with non-coding (or irrelevantly-coding) DNA
It's not risk free. It's just less risky. There are all sorts of elements within the "intergenic" region of genomes which can impact fitness. Duplications in these regions can change the "dosage" of genes or the availability of DNA to be transcribed.
Duplicating these regions is mostly dependent on how tolerate the species is of dosage changes. Humans for example are very intolerant, while plants are very tolerant.
Also, please stop using "junk DNA". It is outdated term ;).
And yes, I'll be dropping "junk DNA" from my vocabulary immediately. I knew some non-coding regions had been found to have uses, but this thread taught me just how extensively relevant it is!
If you're interested in that, I'd look up "Whole Genome Duplications". For instance, corn underwent a WGD about 7 million years ago (IIRC, the number might be off), and we can identify an 'a' and 'b' genome. After these events, most of the duplicated genes will be lost, either through purging selection or drift. However, some will stick around and take on new subset functions (see nkrumm reply).
If you really want to get into the weeds on this, there are many metrics that could, in theory, affect the 'evolvability' of a genome. Higher transposon (or any sort of repetitive element) content can lead to large-scale shuffling of genome structure, such as inversions, duplications, etc. The ecological success of the grass family, for instance, may have it's roots in genome structure.
Now, there's not a whole lot we can say about this because it's not something we have a really good grasp of. It's mostly some hypotheses that are quite difficult to test, but this is some of the most interesting genomic research going on, in my opinion.
All of this is quite difficult to understand, however, as it really depends on epigenetic mechanisms like chromatin state, dicers, RISC, and all that. Figure that out, and you'll realize there's no 'junk' DNA, just various non-coding sequence that's part of this larger-scale genome evolution.
Why is it that plants are so much better than animals at tolerating chromosomal shenanigans? Plants can be highly polyploid, get chromes swapped out, randomly get duplicates, etc., and not seem to suffer massive problems. In a human, you duplicate one chromosome, and you get a very buggy phenotype.
One reason is the absence of sex chromosomes in plants. If humans get a duplicated sex chromosomes things go completely haywire.
Another reason is that due to so many WGDs, plants have ample gene copies, so the copies can freely mutate around or even break, there's always a backup.
Not all of these shenanigans work out, in plants this has the wonderful name 'genomic shock'. In wheat there's a candidate gene (Ph-1) which seems to somehow stabilise these instabilities between chromosome copies. In other species people have been trying for a few years to make polyploids but have been failing because there is no stability mechanism - for example, there are only few papers where people managed to make a Brassica hexaploid, these hexaploids are usually very unstable and have problems generating offspring (see for example http://www.cropj.com/malek_7_9_2013_1375_1382.pdf )
> “The problem is, when you have the same exact words, it’s hard to assemble,” said Lander, just as if jigsaw puzzle pieces show the same exact blue sky.
This is inaccurate, though the fault is with the author, not Lander. When you have genomic repeats, the appropriate analogy is if you had multiple puzzle pieces that had the exact same edges on all sides.
Puzzle pieces that show the same blue sky but with different edges can still be uniquely assembled. But puzzle pieces with identical edges create non-unique but equally satisfactory assembly solutions, and this is the case with the problem of genomic assembly.
A question and a story for any intrepid biologists and geneticists. I'm seeing in other comments that this is fundamentally a problem with how genome sequencing works today.
When I was a movie theatre projectionist, I had a similar problem. Movies have to be assembled onto platters, from a handful of reels which contain 10-20 minute lengths of film. There's about a mile of film per hour, and to assemble it quickly you have a motorized platter and a table with its own motor with dials to control the speed of one or the other. How it should work is you take a center ring - a circle with a gap to wind film through and metal spokes that sit in the platter - and you spin the film on. You put the reels one by one onto this, stopping with each one to cut the footer of one and the header of the next to splice them together with tape.
Well, that's how it's supposed to work. Sometimes, the film tightens up, when moving the film from one platter to another. The center ring spokes might not fit, or might create additional stress.
Well, I went full speed ahead anyway, and when a tiny bit of slack resulted in a sudden jerk on the platter, the center ring popped off, and about an hour of film flew over my hand and into the wall behind me. What happened next is almost impossible to describe. A circular disk of film and metal hitting a solid wall briefly became a vertical column as the tension-less film was forced to go any direction except forward, and the result then splayed itself out on the floor in front of me in a tangled mess a mile long.
Perhaps like an early sequencing machine, my first attempt at recovering this was to look at a piece of the film and determine what part of the movie it was from, and to begin throwing away the parts I thought were from trailers.
They weren't. It was the wrong movie.
I despaired, the film was a mile long and tangled into unbelievable knots. I had to cut it to untangle it. But how?
And it hit me, I knew where a handful of colored markers were, and I began randomly pulling segments of film out from the mess, placing masking tape across a frame, and making 3 lines in 3 different colors. I made dozens of these little loops, and that was how I put it together. The process: cutting, labeling, and placing the segments into new reels made my shift from that night run into the next day's matinee.
This process was painstaking, and that's where my question leads: can we label the ends when we cut up the DNA, like I did with my film? Are we stuck watching chromosomes splash against the wall and shatter into indecipherable pieces?
This is a nice analogy to genome assembly and there are some technologies that try to label DNA before sequencing to aid reconstruction. The one that comes to mind when reading your post is the 10X Genomics system, which acts as a preprocessor before Illumina sequencing. The idea is that you can introduce nucleotide tags into the DNA molecule(s) before sequencing, then use the tags to figure out what short reads came from the same DNA fragment later on. There is a nice video on 10X's website explaining how this works and how it helps:
Haven't paid attention to this area in a while but back in the day I've wondered about the garbage-in garbage-out problem in genome databases. A lot of subsequent sequence assemblies were made on the basis of approximate matching to certain results in the database that have no quality information, and conclusions made on the basis of further approximate matching across datasets. Has anyone seriously worked out how reliable some of these conclusions are? At least physicists spend a lot of time worrying about p-values.
95 comments
[ 2.6 ms ] story [ 164 ms ] thread> “As a matter of truth in advertising, the ‘finished’ sequence isn’t finished,” said Eric Lander, who led the lab at the Whitehead Institute that deciphered more of the genome for the government-funded Human Genome Project than any other. “I always say ‘finished’ is a term of art.”
The phrasing of your parent is arguably dismissive and does little to further substantive discussion. Can one meaningfully discuss "the human genome"? As you point out, each human (with the exception of identical twins) has a unique genome. I think it's reasonable and useful to be able to say "the human genome" as opposed to "the chimpanzee genome" or "the drosophila genome".
If one would rather insist on using the indefinite article, i.e., "a human genome", one can present it without resorting to a rhetorical question replete with scare quotes.
Did you mean each cell will have different gene expression ?
I thought all the cells have the same nucleotide sequences barring maybe occasional mutations (like in cancer)
Assuming each mutation is independent, etc, etc we get a back of the napkin via the binomial distribution with mean = n*p ~ 60 mutations per division.
[1] https://en.wikipedia.org/wiki/Human_genome
[2] https://en.wikipedia.org/wiki/Mutation_rate
However, that is for germline, somatic mutation rates are apparently 10-100x more more common: https://www.nature.com/articles/ncomms15183
These rates no doubt vary by cell type, environment, etc. I was actually thinking something like 1e-6 to 1e-10 mutations/bp/division.
Somatic mutations are just errors in genome duplication in somatic cell lines - i.e. not germ line. These accumulate over time and, yes, can be oncogenic. Many others simply have no consequence at all.
For many organisms, the germ line is 'sequestered' fairly early in development, in part to limit the number of replications needed, and thus errors that accumulate.
Members of a species share most of their genetic material with each other, but there are known areas of high variability between individuals. So to build a specific individual's sequence, it seems like they take the baseline "human genome" data, and overlay the areas that are known to change.
This is true, but only for Insertions/Deletions. The substitution error rate is comparable to Hi-Seq.
So it's good for resequencing but without a reference or decent scaffolds, you're in the dark.
ONT will be much more useful for assembling reference genomes (which can then be polished with short reads), characterizing large scale structural variants, and real time applications like "what pathogen am I infected with?".
If you're doing a standard RNA-seq experiment, short 100bp to 200bp Illumina reads are just dandy for what you need. Most of the time, at least; you could, for example, be interested in different splice variants for a gene family; in that case longer continuous reads are valuable. But then it's no longer a standard RNA-seq.
PacBio and pore-type sequencing is most valuable for polishing reference genome assemblies. For instance, this is a huge issue for crop plants because they have huge, repetitive genomes and are often polyploid. So there's a lot of interest in creating high quality assemblies not just for e.g. corn, but the major research inbreds (B73, Mo17, etc.) as well as private breeding lines. Some of the most valuable crop QTLs (R genes, microRNAs, transposon fragments, etc.) are just 'junk' sequence that plays a regulatory or defense role, so it's super important.
In general, the Illumina data is very high quality and very inexpensive, and works beautifully provided you have a high-quality assembly to work with, and what you're interested in isn't very repetitive. It has many good applications, so I don't see much risk at this point.
If, at some point, longer reads can be done with high fidelity and with similar cost, then there would be an issue. I don't think that's going to happen anytime soon, but I'm sure Illumina is looking at that long-term.
Don't get me wrong, the Minion is amazing, but it can't compare to PacBio in terms of quality... yet.
Substitution error rate was very low, but you should evaluate INDELs specifically, rather than just identity like you did on your git, because it is the major drawback of this technology.
The non-random nature of their error profile is also a problem.
Also, which library prep are you using? I think the ligation kit has historically gotten better yield (not sure if that's true any more).
I'm also extremely put off by how they handle support -- that is via a forum where they do their very best to barely answer anything
I used the flowcell version previous to the current one and got ~2gb of data. I used the ligation kit on genomic extraction of ~10 individuals (crustacean).
or did you think science was a book of unambiguous answers that only the godly scientist may read?
https://news.ycombinator.com/newsguidelines.html
And: "“I’m between agnostic and a little skeptical that these bits will be important for disease, but maybe I’m saying that because we can’t read them,” Lander said.".
> The reason for these gaps is that DNA sequencing machines don’t read genomes like humans read books, from the first word to the last. Instead, they first randomly chop up copies of the 23 pairs of chromosomes, which total some 3 billion “letters,” so the machines aren’t overwhelmed. The resulting chunks contain from 1,000 letters (during the Human Genome Project) to a few hundred (in today’s more advanced sequencing machines). The chunks overlap. Computers match up the overlaps, assembling the chunks into the correct sequence.
> That’s between difficult and impossible to do if the chunks contain lots of repetitive segments, such as TTAATATTAATATTAATA, or TTAATA three times. “The problem is, when you have the same exact words, it’s hard to assemble,” said Lander, just as if jigsaw puzzle pieces show the same exact blue sky.
> In 2004, the genome project reported that there were 341 gaps in the sequence. Most of the gaps — 250 — are in the main part of each chromosome, where genes make the proteins that life runs on. These gaps are tiny. Only a few gaps — 33 at last count — lie in or near each chromosome’s centromere (where the two parts of a chromosome connect) and telomeres (the caps at the end of chromosomes), but these 33 are 10 times as long in total as the 250 gaps.
I recommend reading HN user eggie's comments on this from last week.
https://news.ycombinator.com/item?id=15482439 https://news.ycombinator.com/item?id=15483462
Imagine you have a string of length 3 billion made by randomly choosing from 4 characters. Like this
you get to randomly sample 1 billion[3, page 7] overlapping substrings of length 200[3, page 7] with .1% of the characters randomly changed[3, page 8]. Trying to find the original string from this is technically an undecidable problem. If there's a sequence 400 characters long that repeats multiple times, how could you know if it repeats 5 times or 50 times? (this would be unlikely to happen with random.choices() but DNA isn't random). This is called sequence alignment and it's one of the hard problems in bioinformatics[4].[0] https://docs.python.org/3/library/random.html random.choices() was added in Python 3.6
[1] http://www.biology-pages.info/B/BasePairing.html source for `weights`
[2] https://en.wikipedia.org/wiki/Human_genome source for 3_234_830_000 (python ignores underscores in numbers)
[3] http://sci-hub.io/10.1111/j.1755-0998.2011.03024.x illumina is the most popular producer of genome sequencers
[4] https://news.ycombinator.com/item?id=5123377
On a tablet, I highlight and pull down. It grabs it all and I can then read it. It is a very awful workaround that impacts the site's usability.
The alternative technologies, sometimes called third generation sequencing are nanopore based (PacBio and Oxford Nanopore technologies), they work by pulling long DNA molecules through tiny holes and read the sequence directly, more specifically, they read a couple of base pairs at a time and infer the sequence from the current between the sides of the pores. This is very error prone and the (around 1 in 10?) random errors are corrected by sequencing every part of the genomic area of interest multiple times.
To be clear, most sequencing experiments (in research and in the hospital) are designed to look at variations we can understand and thus the "words" (we call them reads) are aligned against "the" reference genome (as opposed to assembling de novo), which is the genome for as far as we understand it (the last official version is from 2013 but is updated regularly with new parts). Typically, a significant portion of the reads a sequencer produces cannot be mapped to the reference, probably as a results of unmapped regions indeed or perhaps deviations between the patient and the reference genome, this is of course even worse in patients with tumors with unstable genomes. This field still has a lot of challenges to tackle.
Nope, not randomly. Having a random error profile is the best case. In reality these machines suffer from systematic errors, especially the Illumina ones. That means that they always fail to read through certain pieces of DNA, or make the same error every time.
The other thing this article doesnt mention is the idea of coverage. Its not like a jigsaw puzzle with one copy of each tile-- every time a whole genome is sequenced it is done so 30 times over to ensure that the pieces are correct.
Finally, Read length varies from technology to technology. This one claims average 10k base pairs.
http://www.pacb.com/smrt-science/smrt-sequencing/read-length...
I realize that the magnitude of this natural shuffling, even during meiosis, is nowhere near like what happens when sequencing DNA[2], but (assuming I'm correct) it's helpful to realize that this problem of deciphering the "correct" sequence reflects something intrinsic to the operation of the machinery; that this applies even when the relevant macro sequences are all genes that directly code for phenotypes as traditionally understood; and that gene sequencing works precisely because many (most?) genes are relatively tolerant to being spliced in at random locations in the strand. (Which is different than saying all locations--or even any two locations--produce identical results.)
[1] I refrain from saying "complete set of genes" because I imagine for some genes location at both the micro- and macro-scale is everything. AFAIU the term gene is sometimes better understood as more a description of the result of possibly non-local sequences which maintain reproductive affinity than an identification of a specifically adjoining sequence of base pairs.
[2] Or is it?
This is correct, and is a reason why genome informatics has been moving away from representing the genome as a single linear "reference" sequence of 3 billion nucleotides, and towards a graph-based model: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5411762/.
There’s even evidence that this is true, for who knows their audience better than them? You, while making up a a portion of the audience, are not the main portion, or so it seems, unless they don’t understand their audience that well after all... although that seems pretty unlikely
So you have two problems. The first is, you never read anything longer than say 1k bp ever. The second is say your nuclease cuts at GAATTC. You will have great difficulty counting runs of that sequence.
this is the technique https://en.wikipedia.org/wiki/Polyacrylamide_gel_electrophor...
The other takeaway I had, as someone who came to this with a CS and stats background, is that chemistry sucks. None of it is deterministic, and when people say some chemistry does X, what they really mean is it generally does that. Most of the time. Subject to conditions and a reasonable amount of stochasticity. We're built on a pile of random muck hacked together.
And yes, the chemistry that biology takes advantage of means embracing uncertainty in the world. It's all thermodynamic chances, with lots of non-linearity added on top of it all.
Alternative splicing, readthrough regulaton, multiple kinds of ribosomes, post-translational regulation, etc. all produce significant variation outside of the dna itself.
We have about a 0.0001% comprehension of this whole system, how can you possibly be so arrogant as to think you KNOW that certain parts of the sequence are not important or used in _any_ way? Then again, I'm definitely not qualified to comment on the topic.
from TED: https://youtu.be/WFCvkkDSfIU?t=3m39s
A question I've never seen addressed: is one justification for junk DNA to create space for beneficial mutations?
Obviously some mutations are actively harmful, but the reason most mutations are destructive is that they break something already-useful into something inert. That's not a risk with non-coding (or irrelevantly-coding) DNA, which means that in return for spending energy on copying junk DNA, we get space where even barely-useful changes will be net positive.
Is this obvious to experts? Completely stupid for reasons I don't know? Even a possibility worth discussing?
See wikipedia [1] for an overview.
[1] https://en.wikipedia.org/wiki/Gene_duplication#As_an_evoluti...
Yes, gene/genome duplications can help create raw material for adaptation, or be adaptive by their own right.
>That's not a risk with non-coding (or irrelevantly-coding) DNA
It's not risk free. It's just less risky. There are all sorts of elements within the "intergenic" region of genomes which can impact fitness. Duplications in these regions can change the "dosage" of genes or the availability of DNA to be transcribed.
Duplicating these regions is mostly dependent on how tolerate the species is of dosage changes. Humans for example are very intolerant, while plants are very tolerant.
Also, please stop using "junk DNA". It is outdated term ;).
And yes, I'll be dropping "junk DNA" from my vocabulary immediately. I knew some non-coding regions had been found to have uses, but this thread taught me just how extensively relevant it is!
If you're interested in that, I'd look up "Whole Genome Duplications". For instance, corn underwent a WGD about 7 million years ago (IIRC, the number might be off), and we can identify an 'a' and 'b' genome. After these events, most of the duplicated genes will be lost, either through purging selection or drift. However, some will stick around and take on new subset functions (see nkrumm reply).
If you really want to get into the weeds on this, there are many metrics that could, in theory, affect the 'evolvability' of a genome. Higher transposon (or any sort of repetitive element) content can lead to large-scale shuffling of genome structure, such as inversions, duplications, etc. The ecological success of the grass family, for instance, may have it's roots in genome structure.
Now, there's not a whole lot we can say about this because it's not something we have a really good grasp of. It's mostly some hypotheses that are quite difficult to test, but this is some of the most interesting genomic research going on, in my opinion.
All of this is quite difficult to understand, however, as it really depends on epigenetic mechanisms like chromatin state, dicers, RISC, and all that. Figure that out, and you'll realize there's no 'junk' DNA, just various non-coding sequence that's part of this larger-scale genome evolution.
One reason is the absence of sex chromosomes in plants. If humans get a duplicated sex chromosomes things go completely haywire. Another reason is that due to so many WGDs, plants have ample gene copies, so the copies can freely mutate around or even break, there's always a backup.
Not all of these shenanigans work out, in plants this has the wonderful name 'genomic shock'. In wheat there's a candidate gene (Ph-1) which seems to somehow stabilise these instabilities between chromosome copies. In other species people have been trying for a few years to make polyploids but have been failing because there is no stability mechanism - for example, there are only few papers where people managed to make a Brassica hexaploid, these hexaploids are usually very unstable and have problems generating offspring (see for example http://www.cropj.com/malek_7_9_2013_1375_1382.pdf )
This is inaccurate, though the fault is with the author, not Lander. When you have genomic repeats, the appropriate analogy is if you had multiple puzzle pieces that had the exact same edges on all sides.
Puzzle pieces that show the same blue sky but with different edges can still be uniquely assembled. But puzzle pieces with identical edges create non-unique but equally satisfactory assembly solutions, and this is the case with the problem of genomic assembly.
When I was a movie theatre projectionist, I had a similar problem. Movies have to be assembled onto platters, from a handful of reels which contain 10-20 minute lengths of film. There's about a mile of film per hour, and to assemble it quickly you have a motorized platter and a table with its own motor with dials to control the speed of one or the other. How it should work is you take a center ring - a circle with a gap to wind film through and metal spokes that sit in the platter - and you spin the film on. You put the reels one by one onto this, stopping with each one to cut the footer of one and the header of the next to splice them together with tape.
Well, that's how it's supposed to work. Sometimes, the film tightens up, when moving the film from one platter to another. The center ring spokes might not fit, or might create additional stress.
Well, I went full speed ahead anyway, and when a tiny bit of slack resulted in a sudden jerk on the platter, the center ring popped off, and about an hour of film flew over my hand and into the wall behind me. What happened next is almost impossible to describe. A circular disk of film and metal hitting a solid wall briefly became a vertical column as the tension-less film was forced to go any direction except forward, and the result then splayed itself out on the floor in front of me in a tangled mess a mile long.
Perhaps like an early sequencing machine, my first attempt at recovering this was to look at a piece of the film and determine what part of the movie it was from, and to begin throwing away the parts I thought were from trailers.
They weren't. It was the wrong movie.
I despaired, the film was a mile long and tangled into unbelievable knots. I had to cut it to untangle it. But how?
And it hit me, I knew where a handful of colored markers were, and I began randomly pulling segments of film out from the mess, placing masking tape across a frame, and making 3 lines in 3 different colors. I made dozens of these little loops, and that was how I put it together. The process: cutting, labeling, and placing the segments into new reels made my shift from that night run into the next day's matinee.
This process was painstaking, and that's where my question leads: can we label the ends when we cut up the DNA, like I did with my film? Are we stuck watching chromosomes splash against the wall and shatter into indecipherable pieces?
https://www.10xgenomics.com/technology/
I wonder if he said this with a straight face.