Slashdot Mirror
Sequencing a Human Genome In a Week
blackbearnh writes "The Human Genome Project took 13 years to sequence a single human's genetic information in full. At Washington University's Genome Center, they can now do one in a week. But when you're generating that much data, just keeping track of it can become a major challenge. David Dooling is in charge of managing the massive output of the Center's herd of gene sequencing machines, and making it available to researchers inside the Center and around the world. He'll be talking about his work at OSCON, and gave O'Reilly Radar a sense of where the state of the art in genome sequencing is heading. 'Now we can run these instruments. We can generate a lot of data. We can align it to the human reference. We can detect the variance. We can determine which variance exists in one genome versus another genome. Those variances that are cancerous, specific to the cancer genome, we can annotate those and say these are in genes. ... Now the difficulty is following up on all of those and figuring out what they mean for the cancer. ... We know that they exist in the cancer genome, but which ones are drivers and which ones are passengers? ... [F]inding which ones are actually causative is becoming more and more the challenge now.'"
Functions that don't do anything, no comments, worst piece of code ever!
I say we fork and refactor the entire project.
Just store all that data as a chemical compound. Maybe a nucleic acid of some kind? Using two long polymers made of sugars and phosphates? I bet the whole thing could be squeezed into something smaller than the head of a pin!
Typically they sequence every base at least 30 times.
How we know is more important than what we know.
I wondered the same thing, so I asked. From the article: And between two cells, one cell right next to the other, they should be identical copies of each other. But sometimes mistakes are made in the process of copying the DNA. And so some differences may exist. However, we're not at present currently sequencing single cells. We'll collect a host of cells and isolate the DNA from a host of cells. So what you end up is with when you read the sequence out on these things is, essentially, an average of this DNA sequence. Well, I mean it's digital in that eventually you get down to a single piece of DNA. But once you align these things back, if you see 30 reads that all align to the same region of the genome and only one of them has an A at the position and all of the others have a T at that position, you can't say whether that A was actually some small change between one cell and its 99 closest neighbors or whether that was just an error in the sequencing. So it's hard to say cell-to-cell how much difference there is. But, of course, that difference does exist, otherwise that's mutation and that's what eventually leads to cancer and other diseases.
We pissed away $3 billion dollars and 13 years of time, when we could have waited a few more years and got it done in a week, and much, much cheaper. What a waste of time and money that was....
I know I'm being trolled, but you're an idiot. It's pretty obvious that the ability to sequence the genome in a week could only result from techniques developed and information gathered in the original Human Genome project.
I had but a simple dream, to destroy all humans.
It wasn't the computing power that was the holdup, it was the sequencing throughput. Also, as noted in the article, they can do it in a week now partially because they have the completed human genome to use as a template to match things up against. As I analogized in the interview, it's like the difference between putting together a jigsaw puzzle with the cover image available, and doing one without.
What's funny is that there is actually people who think like that. Apparently if we just sit around and wait, things will get better. I call this the dark side of the "invisible hand" of the market.. because it is invisible, people forget how it comes about. In order to get improvement in technology you need a market for that technology. And, typically, you need some loss-leader to create the market in the first place. Government funding serves this purpose well.
How we know is more important than what we know.
Data handling and analysis is becoming a big problem for biologists generally. Techniques like microarray (or exon array) analysis can tell you how strongly a set of genes (tens of thousands, with hundreds of thousands of splice variants) are being expressed under given conditions. But actually handling this data is a nightmare, especially as a lot of biologists ended up there because they love science but aren't great at maths. Given a list of thousands of genes, teasing out the statistically significantly different genes from the noise is only the first step. Then you have to decide what's biologically important (e.g. what's the prime mover and what's just a side-effect), and then you have a list of genes which might have known functions but more likely have just a name or even a tag like "hypothetical ORF #3261", for genes that are predicted by analysis of the genome but have never been proved to actually be expressed. After this, there's the further complication that these techniques only tell you what's going on at the DNA or RNA level. The vast majority of genes only have effects when translated into protein and, perhaps, further modified, meaning that you cant's be sure that the levels you're detecting by the sequencing (DNA level) or expression analysis chips (RNA level) actually reflects what's going on in the cell.
One of the big problems studying expression patterns in cancer specifically is the paucity of samples. The genetic differences between individuals (and tissues within individuals) means there's a lot of noise underlying the "signal" of the putative cancer signatures. This is especially true because there are usually several genetic pathways that a given tissue can take to becoming cancerous: you might only need mutations in a small subset of a long list of genes, which is difficult to spot by sheer data mining. While cancer is very common, each type of cancer is much less so; therefore the paucity of available samples of a given cancer type in a given stage makes reaching statistical significance very difficult. There are some huge projects underway at the moment to collate all cancer labs' samples for meta-analysis, dramatically increasing the statistical power of the studies. A good example of this is the Pancreas Expression Database, which some pacreatic cancer researchers are getting very excited about.
I suppose it's worth noting that the intermediate (raw) data sets can get pretty large. they are actually getting larger as the trend goes towards shorter less informative "reads" that require more of them to recover the connective information and to recover from errors and duplications. However that's a tend that has a stopping point. While more reads is better at some point there is almost no added value from more reads. So at that point that's the maximum amount of data you need to collect. it's won't increase ever. meanwhile hard drive and network speeds will go up factors of ten.
thus the storage issues here are well tolerated at present and soon will become trivial.
Some drink at the fountain of knowledge. Others just gargle.
Next gen sequencing eats up huge amounts of space. Every run on our Illumina Genome Analyzer II machine takes up 4 terabytes of intermediate data, most of which comes from the something like 100,000+ 20 Mb bitmap picture files taken from the flowcells. All that much data is an ass load of work to process. Just today I got a little lazy with my Perl programming and let the program go unsupervised...and it ate up 32 gb of ram and froze up the server. Took redhat 3 full hours to decide it had enough of the swapping and kill the process.
For people not familiar with current generation sequencing machines, they can scan between 30-80 bp reads and use alignment programs to match up the reads to species databases. The reaction/imaging takes 2 days, prep takes about a week, processing images takes another 2 days, alignment takes about 4. The Illumina machine achieves higher throughput than the ABI ones but gives shorter reads; we get about 4 billion nt per run if we do everything right. Keep in mind though, that 4 billion that they mention in the summary is misleading: the read cover distribution is not uniform (ie you do not cover every nucleotide of the human's 3 billion nt genome). To ensure 95%+ coverage, you'd have to use 20-40 runs on the Illumina machine...in other words, about 6-10 months of non-stop work to get a reasonable degree of coverage over the entire human genome (at which point you can use programs to "assemble" the reads in a contiguous genome). WashU is very wealthy so they have quite a few of these machines available to work at any given time.
the main problem these days is that processing all that much data requires a huge amount of computer knowhow (writing software, algorithms, installing software, using other people's poorly documented programs), and a good understanding of statistics and algorithms, especially when it comes to efficiency. Another problem they never mention are artifacts from the chemical protocol; just the other day we found a very unusual anomaly that indicated the first 1/3 of all our reads was absolutely crap (usually only the last few bases are unreliable); turned out our slight modification of the Illumina protocol to tailor it to studying epigenomic effects had quite large effects of the sequencing reactions later on. Even for good reads, a lot of the bases can be suspect so you have to do a huge amount of averaging, filtering, and statistical analysis to make sure your results/graphs are accurate.