Ion Torrent’s chip-based genome sequencer is poised to revolutionize medicine | Hacker News Reader

Ion Torrent’s chip-based genome sequencer is poised to revolutionize medicine

beta.spectrum.ieee.org

37 pointsgourneau45 comments

I am developer at Ion Torrent. If you want to help hack the future of medicine, I am looking for Python and Django hackers to join my team in San Francisco. We are a totally open source shop.

We also need awesome electrical engineers, mechanical engineers, signal processing engineers, CUDA hackers, FPGA developers, statisticians (we use R too), bioinformatics folks, JavaScripters, UI people, distributed systems engineers, documentation and developer evangelists, DevOps and SREs, and testers. Not to mention all the wet lab bio stuff.

email me at josh@gourneau.com if you are interested

Hi Ion Torrent dev. This is a really exciting project.

The article says "decipher a human genome in a few hours, and at the bargain-basement price of $1000".

Does this mean the hardware will cost $1000, or does a complete sequence cost $1000 of consumables/depreciation?

That means the cost of consumables is $1000.

So how does Django relate to gene sequencing?

It helps to think of our machines as fancy scanners. We have to build software to help our users manage their petabytes 'scanned' of data.

We also build developer tools to help the researchers and scientist extend our system with plugins.

One cool thing is that all of our software is open source. Check out the Django project I work on here.

https://github.com/iontorrent/TS/tree/master/dbReports

Are you now open to receiving pull requests - there seemed to be a legal problem accepting a patch on the above, and if so what issues do you need fixed :-)

Hi gourneau, I'm a grad student working on the hardware side of nanopore sequencing. Do you know anything about opportunities in nanofabrication/system design? Or does Dr. Rothberg come up with these things all on his own?

If I can telcommute I have some free time.

Incidentally, anyone who wants to play with some genome processing is encouraged to download pygr[1]. You can connect to worldbase[2] and explore all the different genomes they have on their server using Python.

[1] http://code.google.com/p/pygr/ [2] http://code.google.com/p/pygr/wiki/ServingDataUsingWorldbase

How big is a fully sequenced genome, as a file?

http://hgdownload.cse.ucsc.edu/goldenPath/hg19/bigZips/

~800MB for human genome

That is roughly the size needed to represent 3.3 billion ACTGs , however sequencer output is much larger since each base is sequenced many times, and accompanying metadata (quality information) is also included. File sizes for current sequencer output is on the order of 100-300gb compressed. It's big enough to create real roadblocks when you are trying to do analysis.

It can also end up well below 800MB per person if you encode it as known sequences of DNA + tiny new random variations.

True, but in order to do that you have to know what the new random variations are. Output from existing sequencers, as well as the current crop of algorithms, are not good enough to do that with high confidence. Thus, people tend to keep everything. You can get some gain by doing reference-based compression, but you're still looking at on the order of 100s of gb per sample.

When compressed relative to a reference genome, you can get a ~2.9Gbp human genome down to ~3.3MiB of data[1]. Sequences are actually pretty compressible in general, owing to conservation of functional domains and repeats.

Data from next-gen sequencers have many small overlapping segments though, so the raw output takes up MUCH more space.

By the way, for anyone looking for more information about some of the methods currently being used for sequencing (Illumina, 454), check out these videos from the WUSTL Genome Center:

http://gep.wustl.edu/curriculum/course_materials_WU/introduc...

They don't cover Ion Torrent, which is unfortunate because it is way cool. gourneau can probably point to a better video, but here are a couple for Ion Torrent:

http://www.youtube.com/watch?v=W3jPTBU9vF8

http://www.youtube.com/watch?v=yVf2295JqUg

...and the Nature paper:

http://www.nature.com/nature/journal/v475/n7356/full/nature1...

1. http://dhruvbird.com/genome_compression.pdf

How do you deal with patents? In my view, an open source shop is not compatible with holding patents, but the space is very heavily patent-encumbered. How do you resolve the conflict?

I emailed you about meeting up at Pycon but I never heard back.

One point worth mentioning. Ion Torrent is one of several companies that develops high-throughput or 'nextgen' sequencing platforms 454 and Illumina being the other big players right now. Each technology has a few important nuances that make them better or worse for a particular project. Some of the major things one looks at are cost of the device and library construction, error rates due to PCR amplification steps or nucleotide calling, library sizes, and maximum read length but there are many more. In addition, pore technologies will definitely become important in the near future because they enable single molecule experiments which are currently impossible without amplification using the above platforms.

Ion torrent appears to be the same platform as 454, but with detection via chemical means (ISFETs) rather than optical transduction. One of the big drawbacks of the 454 technology that doesn't get talked about a lot is that the immulsion PCR/microbeads require a lot of preparation before the actual sequencing run. On the 454, I believe the sequencing time was ~8hrs, but the preparation time (at least at the facility where I was involved) typically took more than 8 hrs. In terms of actual sequencing time, it's the best system. But as you said, read length is a major issue with all of the current "nextgen" systems that should be resolved with nanopore sequencing.

From a clinical / experimental standpoint there's very little difference between 4,8,16 hours. It's all essentially 'next day' right now because common practice is to send the sample off to a genomics core and they drop the data on a server for you after they've run it through their pipeline. Many experiments require their own library construction but clinical pipelines generally don't and again, the machine prep time is not the bottleneck it's deciding what to include and how to lay things out. From what I understand 454 accommodates the longest read lengths right now but this seems to change month to month, also from what I understand 454 has a lot of trouble with short length libraries.

PacBio has average read length now of about 5000 bp whereas 454 is still only about 1000 bp.

For $50, I had 23andme read my genotype for a pretty significant number of already studied genes, but far from my complete genome.

How much of a difference does it make to the end user whether their complete genome is sequenced, compared to 23andme's model?

There's a lot of differences, especially because 23andme just checks the "different" nucleotides for some specific genes. Having your full genome sequenced would mean that we would know every single nucleotide from you, so things we still don't know are important might become vital in the future.

"Having your full genome sequenced would mean that we would know every single nucleotide from you"

Unless he's a genetic mosaic, also known as the "reality is hard to squeeze into our all too simplistic data model" problem.

23andme is babyish relative to what's coming. Even exome sequencing does not provide the whole story. There are so many conditionals, so many sequences important in regulation, so much of the "dark" genome that actually codes for RNA. Sure, some SNPs are highly predictive, but there are compensatory mutations and regulatory gene circuits. Everything works together and must be weighed together. Even with whole genome we still need epigenetics and we need time-lapse data and sampled location specific (cell and organ system) genomic data to really produce clinically relevant predictions.

Interpretation is an indefinite field - statisticians and econometricians will be highly valued.

We hope as more data becomes available, the field will become more and more definite. We will use indefinite mathematics in genomic interpretation until we have near perfect information.

I can't find the reference but micro-RNA as short as two base pairs has been shown to affect gene expression. Nonlinearity may seriously inhibit meaningful analysis.

It doesn't make a whole lot of difference at the moment; at present the main utility is related to genealogy rather than health, which the current genotyping chips are adequate for. Until our ability to interpret whole genomes progresses beyond its current rudimentary level, there isn't much more for you to get excited about.

In any case, you just have to wait a few more years and you'll be able to get (the non-junk bits of) your genome sequenced for close to $50 anyway.

Some interesting structural differences are not captured by 23andMe nor even with technology like Ion Torrent, for example repeat sequences related to ALS, Huntington's, etc. To get the full picture ultimately you'll need a full sequence with longer read lengths to span the structural variants, and you'll also want the methylation/epigenetic markers which convey which genes are turned on or turned off. Just knowing you have the gene doesn't mean nearly as much without knowing if it is turned on!

The cost comes from the continuous washing and waves of nucleotide additions (reagent costs). Illumina's technology suffers the same difficulty. The promise of PacBio's technology is that it does not require this continuous washing and staggering of nucleotide additions (unfortunately PacBio was not able to translate its technology to high-throughput high-accuracy or manufacture properly its version of "chips" [SMRT cells] for Single Molecule Real-time Sequencing). I'm afraid the cost will not come down significantly for iontorrent or illumina without a fundamental improvement in reagent utilization techniques or reduction in reagent costs.

I agree PacBio doesn't have the high-throughput of Illumina or similar amplification approaches yet, but from last month's AGBT conference papers it's clear that it's now more accurate than other systems with typical coverage (no bias, covers repeats, Q50-60, etc.)

I'm all in for technological progress and advanced medicine but how long before Gattaca type scenarios become real? Will companies start screening candidate employees? will this split the world in classes of good genes and bad genes? how about dating? will you future wife to be check your genome before saying YES?? its a future the scares me.

The non-beta-site version of the article is here: http://spectrum.ieee.org/biomedical/devices/the-gene-machine...

Forget sequencing - DNA synthesis and designing of new genetic programs/apps (new creatures) using synthetic biology and software like Genome Compiler (genomecompiler.com) is the interesting future.

"Forget reading, we have the ability to write"

Sequencing is critical to quality control in synthetic biology. Better/cheaper/faster sequencing = better/cheaper/faster synthesis. It's like chocolate and peanut butter.

Problem is there is still the "genome-clinic" gap.

Genomics in general has failed to live up to its promises to revolutionize medicine. Why will a cheaper version of the same thing fare significantly better?

I struggle to believe this is true - as a simple one the different cancers that can be identified by gene sequencing tell doctors which treatments will be more or at all effective even when physical symptoms look similar.

There are i am sure many more examples - what do you expect - a couple of flashing LEDs and hey presto cured?

Really? By what standard are you holding this revolution? From my perspective, the field is growing rapidly and I believe it will eventually take off.

See, e.g., http://pipeline.corante.com/archives/2010/06/18/the_economic...

Don't confuse value with economic activity. Painkillers are a perfect example of something that's cheap because of manufacturing costs but whose impact on use is extreme.

It does not show up as just one thing.

There are several 'wonder' cancer drugs for example that came from analyzing DNA and people know to use them by doing DNA sequencing. It's also helped 'cure' several rare but debilitating diseases. Honestly, it's rapidly approaching the value of antibiotics which clearly was a major revolution.

Anyway, the reason why full DNA sequencing is so important has a lot to do with research. When you know what some sequence does then you can look for that. But, being able to build a database of say 100,000 peoples full DNA sequence for say a billion or less let's you look for those sequences far more easily.

Approaching the value of antibiotics? That's a pretty bold claim, and it seems entirely unsupported.

Reply on news.ycombinator.com