Wednesday, August 31, 2011

Will Max-Seq Gain Traction?

      At the beginning of the month, I had dinner with J Adams from Azco Biotech and some friends/colleagues of his and talked over the Max-Seq.   And yes, I did let them pick up the bill -- J wouldn't let me pay for myself.  In Sequence did a nice piece on it the next week, so again I've blown an opportunity to scoop them.  Then somehow, between vacations and other distractions, this piece was stuck in blogger limbo.  But, there are some details I don't see there and some color I think worth adding.
     Azco's main business is to service synthesizers and sequencers and had previously been involved in the servicing of the Polonator.  At first glance, the Max-Seq starts sounding like Polonator 2: designed by the same company (Dover) and much of the supporting molecular biology has been developed by George Church's and Jeremy Edwards' labs.  One big change is that Max-Seq is designed for Intelligent Biosystems' reversible terminator chemistry.  It also turns out that only a little bit of the hardware is truly shared with the Polonator; the fluidics have been completely replumbed and much else as well.  A late change, according to Adams in his conversation with me, is the inclusion of a higher resolution (5Mpixel) and fast CCD camera, which enables substantially faster scanning of the flowcells.

     It is an unusual way to launch a sequencer.  The big players in the field, Illumina, Life and Roche/454, at least build a facade that everything is done in a single shop.  The Japanese earthquake has underscored some of the artifice of this -- Hitachi is actually building the 5500XL -- and it is generally assumed that various well-known molecular biology shops provide the enzymes to be repackaged and branded, but it all looks like one big sales machine.

   In contrast, here the responsibilities are very visibly divided.  Dover makes the instruments and then ships them to IBS, which tests them for performance to spec.  Azco handles the actual delivery to the user, repairs, training and serves as a single point-of-contact for purchasing reagents.  No attempt is made to hide this; if the three companies can be as smooth as Tinker to Evers to Chance, then nobody will mind.

  The chemistry is a particular class of reversible terminators which IBS licensed from Columbia University about four years ago.  One twist the IS didn't report on is that the terminator solutions contain a minority of fluorescently labeled terminators with a majority of unlabeled ones.  This patented approach is intended to give better repetitive yield; polymerases dislike the bulky fluorophores and are more likely to leave chains unextended.  Also, it was explained to me that with base-linked fluorophores, removal of the fluor always leaves some sort of molecular scar, a bit of carbon chain decorating the base.  This can interfere with later extension.  Anything interfering with chain extension can lead to dephasing, which is the death of signal for all clonal sequencing strategies.
   
    At this time, any long benefits of the strategy aren't going to be visible, as the initial reagent kits are for 35bp or 55bp reads.  I was told you could string multiple of these together and the possibility of very long reads was hinted at.  However, this crowd remained unconvinced that long reads are where attention should be paid, and were similarly blase around the lack of paired ends.  The target markets they envision are first replacing arrays for expression profiling and next resequencing (both targeted and whole genome) and they see piling up data cheaply as the route to conquering those.  

   I raised a number of areas where long reads, or paired reads, really pay off, but most were brushed aside.  De novo genome sequencing is interesting, but not where they see a large market.  This isn't an instrument designed for amplicon sequencing.  Three that are harder to bat down are alternative splicing discovery in RNA-Seq (a target) and fusion transcript discovery (again, RNA-Seq) and structural variant discovery in genomic sequencing.  I'd tack onto that one more I've thought up since our meeting, discovering deletions which are larger than a few base, such as the 15 base deletions in EGFR that show up in lung cancer.  It's not that you can't detect any of these with short reads, but algorithmically they are quite challenging when there is so little to go on.

    On the cost side, the instrument is "below $300K", which I would interpret (in the absence of further information) as $299,999.99.  That's a serious chunk of change for small labs and not radically different than a GAIIx or SOLiD 5500XL.  However, the argued advantage is in reagent costs: as it was put to me, imagine running a whole flowcell for the cost of a single Illumina lane (I'm not sure that's a precisely fair comparison, but it's in the right ballpark).  Flowcells are reusuable with a simple alkali wash; it was claimed until you physically crack the window they'll keep producing.

   A big upcoming change is in the template prep.  Currently, the instrument uses beads and emulsion PCR.  In the near future, rolling circle amplification ("rolonies") will be, well, rolled out.  Rolonies offer a much simpler and less messy template prep and the potential to go for very great densities, plus the problems of null and polyclonal beads are essentially eliminated.  So the vision is getting HiSeq-like numbers of reads from a GAIIx-priced instrument for a bit north of MiSeq reagent costs.   One drawback of rolonies, though, is that there isn't a clear way to perform paired-end sequencing on them.  The Church lab is exploring this, but that may not be available for some time. 

   Of course, Illumina offers very long reads and paired ones as well.  While the Max-Seq team is confident the chemistry will support long reads, the paired ends will be challenging with rolonies.  That's being hammered at in the Church Lab, but it does take some imagination to see a route.  
  
   How good is the data?  An E.coli 35bp dataset is available on the website and a human exome set is available on request.  I ran the E.coli dataset and generated on of my now notorious plots on quality (below); the take-home is that whoever is responsible for the base-caller (IBS? Church lab? Edwards lab?) has some calibration work to do.  While the caller routinely spits out phred scores of 40, the real data reliability is well south of there.  This does point out one value of MaxSeq to IBS -- they will get some additional testing of their chemistry in the field prior to launch of their benchtop PinPoint Mini sequencer.

   Having thought off-and-on about this for much of a month, I've finally realized some things I should have said at dinner.  The machine sounds good, but will be challenging to market given the muscle of the competition (which J clearly sees as a challenge; "David vs. Goliath" was definitely brought out).  Given this, it won't be any easier if the machine is viewed as limited in any way.  $300K for a machine that will do everything would be a much easier sell.  So even if this group isn't very interested in longer reads (especially longer reads in place of paired ends), it will be important for driving sales.

   The other point I tried to make, and will attempt again, is that an increasingly important route to acceptance is to have commercial service providers on board.  Some of this is try-before-you-buy, but I'm hearing increasingly of companies that just don't want to deal with the high capital costs, roller coaster changes and employee training to have their own sequencer.   Yes, Edwards' lab at UNM will be providing services, but for commercial outfits there is a certain level of concern that academic labs may not always be committed to maintaining timelines.  One of the other members of our party was from a service provider that is contemplating launching MaxSeq services, so perhaps there is hope here.

   Personally, I hope MaxSeq survives.  Not just because they bought me dinner, but the more players in the field the better.  Competition will drive prices down and performance up.  

3 comments:

Shawn Baker said...

Great post, Keith. They've stated in one of the GenomeWeb articles that the system cost is ~10% below that of a GAIIx, which would put it at about $270k. The operating cost (on a per Gb basis) appears to be less than 2/3rd that of the HiSeq2000. A $2550 run (reagents/flow cell only) should produce 132 Gb of data, or a little over $19/Gb. Assuming 30X coverage, this gets the human resequencing reagent cost down to under $1800. A $20k HiSeq2000 run should produce 600Gb, or $33/Gb (or $3k for a human genome). However, the 132 Gb run for the MAX-Seq assumes a PE run (which their literature states is possible with the bead method).

I think the biggest challenge (as you've pointed out) will be for this team of companies to provide a seamless user experience and a high level of professional support.

FYI, I've placed a summary of the MAX-Seq over on the BlueSEQ NGS Knowledge Bank (http://knowledgebank.blueseq.com/intelligent-biosystems/)

Anonymous said...

It looks like these guys have literally copy/pasted images directly from the Complete Genomics webiste onto their own website:

http://www.azcobiotech.com/reagents/libkits.php
and
http://www.azcobiotech.com/reagents/seqkits.php

compared to:
http://www.completegenomics.com/services/technology/details/

Unless Complete also lifted these figures from a public source, I would say this is pretty strange. It also makes you wonder what IP-related land mines are in store.

Anonymous said...

Unfortunately, the reality is that you just can't create a business model froma single product. Frankly, this can only be the child of academics who have little education on the reality running for profit... If its not wallmart-able, it won't sell... sad bu true.