[pmiguel]

Surely anything beginning with 5 G's won't be registered as a cluster. So bubbles wouldn't contain any clusters either - so you don't get any data. I would expect that there is some algorithm there to detect when something is a G and when it's an intermittent bubble. G's would be localised areas with no signal, bubbles would cover much larger areas.

I would bet no such checking is done and bubble covering a cluster get called as "G's".

Alternatively, it is technically possible to eliminate or drastically decrease the number of bubbles that traffic through the flowcell. Illumina has never seemed the least bit interested in doing this, that I could tell. But maybe they have sufficient impetus here to deal with this issue.

--
Phillip

[ymc]

On our blog we've given our take on Illumina's "$1000" genome and why it still isn't available for most people.

I think it makes sense commercially for them to market it like that in the beginning. Over time, I think lesser players should be able to buy one instead of ten. Application to other species and non-WGS uses should also be available given time.

[ymc]

I think the three buyers of HiSeq X Ten so far probably has a lot of cancer genomes they want to sequence. Now they can go 300x easily. They may also want to run 1000x or more to study tumor heterogeneity.

[snetmcom]

There's only so many institutions that can afford/support a Hiseq X. I imagine they were sold so quickly because they were designed specifically for those customers. They probably helped set the spec's of the system.

[ymc]

What is the reagent cost for NextSeq 500 High output mode for 2x150 and 2x75? Anyone knows? Thanks

[SNPsaurus]

Here are some prices I've heard:
FC-404-1002 | NextSeq™ 500 High Output Kit (150 cycles)
List Price (USD):$2,500.00

FC-404-1004 | NextSeq™ 500 High Output Kit (300 cycles)
List Price (USD):$4,000.00

FC-404-1005 | NextSeq™ 500 High Output Kit (75 cycles)
List Price (USD):$1,300.00

[Genohub]

Really nice piece on the computation costs of the HiSeq X Ten:

http://glennklockwood.blogspot.co.uk/2014/01/the-1000-genome-computational.html?showComment=1389835567115

[ymc]

Really nice piece on the computation costs of the HiSeq X Ten:

http://glennklockwood.blogspot.co.uk...=1389835567115

But BGI's 128 HiSeq 2000s have roughly the same output as 10 HiSeq X.

So if they could handle it several years ago, I think the new HiSeq X owners should be ok. It is not like they ordered 100 each this time around.

[ymc]

Here are some prices I've heard:
FC-404-1002 | NextSeq™ 500 High Output Kit (150 cycles)
List Price (USD):$2,500.00

FC-404-1004 | NextSeq™ 500 High Output Kit (300 cycles)
List Price (USD):$4,000.00

FC-404-1005 | NextSeq™ 500 High Output Kit (75 cycles)
List Price (USD):$1,300.00

For NextSeq High Output PE150, assuming two years shelf life, there can be 365*2*24/29 = 604 run per machine. The per run equipment cost is $250,000/604= $414. Each run costs $4,000. then for a total of 120Gbp output, the cost is $36.78/Gbp.

From a # of read pairs POV, for PE75, 365*2*24/18 = 973 runs per machine. $250,000/973 = $257 equipment cost. $2,500 to run two flow cells. Then for a total of 400M read pairs = $6.89/M read pairs

For HiSeq Rapid PE150, $750,000/(2*365*2*24/39) = $834.76 equipment cost. $7,800 to run two flow cells. Then for an output of 180Gbp, the cost is $47.97/Gbp.

For HiSeq Rapid PE100, $750,000/2*365*2*24/27) = $577.9 equipment cost. $5,800 to run two flow cells. For an output of 600M read pairs = $10.63/M read pairs.

For MiSeq V3 PE300, $100,000/(365*2*24/65) = $371 equipment cost. $1,400 per run. For an output of 15Gbp, the cost is $118/Gbp

For MiSeq V3 PE75, $100,000/(365*2*24/24) = $137 equipment cost. $800 per run. For an output of 25M read pairs, $37.48/M read pairs.

So the conclusion is NextSeq 500 makes sense as expected....

[TonyBrooks]

I would bet no such checking is done and bubble covering a cluster get called as "G's".

Alternatively, it is technically possible to eliminate or drastically decrease the number of bubbles that traffic through the flowcell. Illumina has never seemed the least bit interested in doing this, that I could tell. But maybe they have sufficient impetus here to deal with this issue.

--
Phillip

We''ll have to wait and see. I have yet to see a bubble on the MiSeq, (although we don't look for one, we have yet to see any spike in the per base N plot of fastqc that we see with HiSeq BMS). So they may have learnt a couple of tricks (like embedding the gaskets on the flow cell).

[pmiguel]

We''ll have to wait and see. I have yet to see a bubble on the MiSeq, (although we don't look for one, we have yet to see any spike in the per base N plot of fastqc that we see with HiSeq BMS). So they may have learnt a couple of tricks (like embedding the gaskets on the flow cell).

Alright, different instrument. Probably won't have the issues associated with the HiSeq (bubbles.)

Let's go back to cluster recognition. Looks like 5 initial G's and the cluster will not be seen at all. But, as you mention, that would be less than 0.1% of the amplicons. So the bias would be unlikely to cause issues.

How about 4G's? Do you really think the software would be confident enough of a cluster which only was visible in a single cycle out of the first 5?

How about 3G's?

This whole inferring signal from a lack of signal makes me twitchy...

--
Phillip

[TonyBrooks]

Alright, different instrument. Probably won't have the issues associated with the HiSeq (bubbles.)

Let's go back to cluster recognition. Looks like 5 initial G's and the cluster will not be seen at all. But, as you mention, that would be less than 0.1% of the amplicons. So the bias would be unlikely to cause issues.

How about 4G's? Do you really think the software would be confident enough of a cluster which only was visible in a single cycle out of the first 5?

How about 3G's?

This whole inferring signal from a lack of signal makes me twitchy...

--
Phillip

I guess that could be why cluster density is so low.

[GW_OK]

So I've been watching Jay Flatley's taped talk to investors from yesterday and I've taken some screenshots and notes for your enjoyment:


Better chemistry including new cluster method, better scan mix, speeding up the paired end turnaround chemistries, new custom camera that can do *bi-directional* scanning (allows for 6x speedup), higher powered lasers, new computer system (bye-bye Dell tower) and re-written software


Remember when Ion Torrent said they were going to use chip foundries to change the sequencing world? Wafer-scale flowcell generation for patterned FCs, 12 FC/wafer.


400nm feature size. Top and bottom of FC.


Example patterned FC picture.


700nm center-to-center distance for 400nm-sized features. Plenty of headspace to pack more and more in.


Seeding the current FCs is a problem b/c DNA binds randomly and you can get polyclonal clusters (Poisson distribution). Current size of clusters and camera power can discern polyclonal clusters fairly well, but with 400nm features there's no way to tell them apart.


New cluster chemistry has simultaneous seeding and amplification. Instantaneous amplification of the template as soon as it sits on a cluster site prevents two templates from seeding the same feature via competition. 80% monoclonal features vs 40% from regular FCs.


"Typical" data qualities from a HiseqX, ~120Gbp/lane. 1.8Tbp/run is conservative. 36x human genome/lane.

System is *licensed only for human WGS*, so that's how they keep you from running other stuff.

[GW_OK]

More from the same presentation, NextSeq500:



Larger than normal flowcell. Like, the size of your hand.


2-dye chemistry.


LED light sources and cellphone cameras. Low cost, light weight, no alignment needed on install.


Fluidics module. "Valve and stagger" technology allows for chemistry and imaging at the same time.


2 different flowcells.


Price per Gbp

[BBoy]

Illumina controls the clustering adapters and sequencing primers. It should be trivial to ensure no Gs at the end of the cluster.