DNA sequencing progress has stagnated | Hacker News Reader

DNA sequencing progress has stagnated

twitter.com

27 pointsdbcooper19 comments

Why would sequencing follow Moore's law when it isn't really related to computing? Sequencing technology is limited by our knowledge of chemistry, physics, and mechanical engineering to apply the concepts. None of which is doubling every 2 years.

There are 3rd gen sequencing techs that provide much longer reads than previous tech like Illumina but they also get much higher rate of erroneous reads.

Another point is current sequencing tech is enough for tasks that were just unfeasible to do couple of years ago. Such as HiC (Extracting 3D structure of chromosomes with sequencing) or more recent applications like DNA Microscopy, where you can construct a very high resolution visual image of a cell just by using sequencing.

Our usage of genomic data isn't really limited by sequencing tech since DNA is such a complex structure that we don't even fully understand the data we currently have.

This is exactly right. The Moore's law slide is a cute comparison that has been used to try and demonstrate that sequencing volume has outpaced our computational ability in one context or another. Variously, it is used as commentary on computers' ability to keep up, our need for better storage , our lack of trained professionals, and so on.

Dollars/Mb is not a super-meaningful statistic anyway. If you include quality/value (either as raw base confidence or the value of that base in the context of longer read lengths), the Dollar/Base/Value number has likely kept falling. It's just hard to quantify the "Value" as a number.

As somebody who has worked in genomics for decades, I don't really see progress in DNA sequencing as being motivated by extrinsic factors. Put another way, there really isn't enough scientific discovery being driven by new sequence data to justify the spending. Many articles have been written claiming that sequencing is going to revolutionize medicine. It's been a very useful research tool, and some genomics has been very valuable medically, but it doesn't live up to the hype.

To add to this, I worked in melanoma genomics for 2 years. Off the top of my head I cannot think of a single cancer drug on the market who's target was discovered through NGS.

It was great as a research tool. But I have yet to see it improve patient outcomes

It's not quite _discovering_ a target from NGS, but lots of existing cancer drugs are targeted based on genotyping tumours. Breast cancer is the classic one, where, say, if you can quickly determine if your sample is Her2+ your treatment plan drastically changes. Now, getting enough cancer cells in your tumour sample to reliably genotype at that stage is a separate hard problem.

The key problem is that DNA sequencing for individuals is not useful (yet) for that individual. Right now DNA sequencing is useful for scientific research and for industrial research. But the current price is basically fine for that; it isn't anywhere near the main component of the cost of research.

If there were a reason to get the DNA of each American sequenced, the price would go down drastically. Like if there was any real medical usefulness of having your DNA sequenced. But right now there isn't.

It's kind of like the Apple I era of personal computing. It's finally cheap enough that individuals can get this technology themselves. There just isn't really a great reason to, besides interested hobbyists.

23andme will let you download a zip file of your sequence data. But if I recall the data had only about 4 million nucleotides. So I assumed that I need to get their "baseline" human full genome somewhere and my file is just the difference.

But the above assumption can't be correct given that it still costs $1000 for a full sequence according the the parent article. Can anyone clarify ?

23andMe ran a microarray analysis, which probes your genome at a specific set of locations.

If you want to know how your genome differs versus what's called the "reference genome," then you'd need a whole-genome sequence. That process shards up your genome, sequences it, aligns the pieces back to the human reference, and then calculates a "consensus" that represents the software's best guess as to how your genome relates to the reference.

Then they could provide you with a diff of the consensus with respect to the reference, which would probably be distributed to you in a VCF file (variant call format).

This process is the one that costs more money versus 23andMe's $99.

23andMe only tests for known relatively common mutations (single nucleotide polymorphisms, or SNPs) that have been identified by non-profit large scale sequencing projects done in the past (basically public data now, see dbSNP for instance). You will almost certainly have mutations that are not covered by their panel, so it is not a matter of comparing your 23andMe results to the "baseline" (or "reference genome"). Their panel costs in the hundreds of dollars to run since it is so focused (at least 10x less than a full "exome" test, which would look at the full sequence of all known genes).

Moore's Law in context is doubling every 2 years? This isn't general news; you'd need to be very clued in to understanding what the problem is. Doubling every 2 years is not the natural state of things. That sort of growth rate is hard to maintain even from truly trivial starting points (my favorite example is the wheat & chessboard parable [0]).

The news is processes that don't stagnate fairly quickly after getting past the low hanging fruit of research and development. Processors have maintained an exponential growth rate for ~50 years and that has completely revolutionised every aspect of society and realistically probably our relationship with the world at large. We don't expect things to do that, and although it is taking a lot longer than everyone expected at some point the transistor doubling will have to end.

[0] https://en.wikipedia.org/wiki/Wheat_and_chessboard_problem

In order for the public to really want their genes sequenced,there will have to be privacy in place. Nobody should have access to the data. Just deliver it and provide analysis tools. Even then, it something you only need to do once. Demand won't get higher than that, so demand isn't really going to be a big price driver.

Yep. I refuse to have my DNA sequenced by any of these direct to consumer businesses that think they're going to be the next platforms and start making money on the back end.

> something you only need to do once.

Maybe for the actual sequencing, but unless consumers get access to the raw data, then the analysis tools are not great. See: https://www.cbc.ca/news/technology/dna-ancestry-kits-twins-m...

Maybe divert resources DNA synthesis? Sequencing is very cheap per base, especially compared to a few years ago

This doesn't take the increase of read lengths into account, which also is of value, not only pure read depth.

It does... the graph is price per megabase, not per read.

(or at least, it's supposed to)

But, I don't think the plot is all that accurate anymore though... given the cost/GB and raw capacity of the newer Novaseq's, the cost per individual experiment is just as much a function of how multiplexed the machine can run.

Read length improves read quality, especially for sequences with lots of repetition. https://news.ycombinator.com/item?id=15534325

Mappable sequence isn't what is being measured here... it's just $/Mbase

Quality may have improved though.

Reply on news.ycombinator.com