Thursday, June 23, 2022

AGBT 2022: Overhanging Questions

AGBT broke up a couple of weeks ago and I've failed to write anything here so far.  It was frustrating not attending, but not registering for a meeting in February seemed prudent given the pattern of COVID waves - I hadn't considered (nor would have wanted to bank on) AGBT organizers reacting so well and rescheduling the meeting.  It sounds like a number of attendees did catch the virus at the meeting -- though I'm presumably still quite protected by my infection a month earlier.  Anyways, I'm going to organize this around one to two questions that hover in my head for the different sequencing providers.  AGBT also had a strong spatial angle, but I feel ill-equipped to cover that in the absence of being on the scene -- I don't work with spatial data and so don't have a deep feel for it.  As always, please flag me here or on Twitter or by email for any errors I made -- or any juicy sequencing company gossip you wish to share!

Element Biosciences: When Will Partner Kits Show Up?

Element Biosciences was a Silver sponsor of the meeting and seemed to be well received.  On the sequencer price vs. price per gigabase (or millions of reads) plots that have sprung up, Element's AVITI occupies a sweet spot of offering significant performance at both lower purchase price and lower cost per unit of sequence data than their similar competitors. AGBT marked the formal commercial launch of the system, with early access users discussing their experiences.  Element also announced a 2x75 kit to launch late this year to enable faster and cheaper counting applications.

A big part of Element Biosciences unveil in March was a flurry of announcements of collaborations with major library prep vendors.  This would enable them to cover a wide range of popular library preparation types, reducing a significant barrier of "but it can't run my experiment" responses.  But poking around on some of the largest partner's websites, Watchmaker Genomics, Roche/Kapa and NEB, did not show any kits actually able to be purchased -- only on the Watchmaker Site did "AVITI" as a search term yield anything relevant at all.  

Until the kits are shipping, users wanting those may be hesitant to invest in AVITI.  Yes, you can convert Illumina libraries, but with a penalty in terms of amount of library for same amount of input.  Element's sequencer does require circular templates rather than linear ones, so adapting kits for AVITI isn't going to just be a simple matter of swapping out adapter oligos.

Singular Genomics: Will Speed & Convenience Sell?

Singular also had a significant presence at the conference as a "contributing sponsor".  Singular looks more expensive for both instrument and 

 Singular's most notable announcement is a MaxReads kit for 4 billion shorter (30-100 basepair) reads to support counting applications.  Few details were released, but one speculation is that this flowcell has much tighter pitch to enable more but tinier clusters that peter out sooner than their standard clusters.

Singular also announced a partnership with Olink to use MaxReads to support Olink's oligo-linked antibody proteomics technology.  Since Illumina recently partnered with Olink rival SomaLogic, partnering with another sequencing tech company made good sense.  Presumably this is a "most favored nation" sort of deal -- Olink isn't going to cut Illumina users off!

Genapsys A No Show:  Will It Pop Up Again?
Genapsys reportedly (I have my spies!) had zero presence at AGBT this year, contrasting with their booth back at the 2020 meeting held as the pandemic was first rising.  Genapsys originally unveiled at AGBT 2014, then basically went silent for many years then later popped up again.  

But Genapsys' website looks like that of a development-stage company, not someone actually selling product.  As I've commented before, they utterly missed the boat on SARS-CoV-2 sequencing, which would have been a good fit for them - in contrast Oxford Nanopore punched far above their weight in terms of sequencing market share, and even Ion Torrent was a significant contributor.

Genapsys with a high density chip would have been a fierce competitor, given the tiny capital cost and modest laboratory footprint of the devices yet data quality as good or better than Ion Torrent and a modest running cost.  I don't hold much hope in it, but if Genapsys never returns it would be so valuable to the community to get an analysis of where they crashed on the technical side.  Was it a problem with getting data off the devices as is rumored for Ion (though I've had someone in the know hint that this isn't necessarily the story)?  Did the ability to read signal break down as the spots got close together?  

Will Ion Torrent Ever Be Interesting Again?

Speaking of Ion Torrent, can I get away with just not mentioning them anymore like I do with SeqLL?  Ion has a focus on their modest customer base, but do they ever actually convince someone new to buy the platform?

Will Oxford Nanopore Ever Return to AGBT?

After their famous 2012 talk, ONT has not had a formal presence at AGBT.  In private they'll compare attending to visiting Mordor.  Their social media staff does actively promote the talks and posters at AGBT, but that's it.  

But AGBT is a spectacular concentration of sequencing customers, sequencing ecosystem providers and investors, and as a commercial company how can ONT justify passing on the opportunity?  Of course, since ONT really doesn't do deals with other companies in the ecosystem - their formal relationships with NVIDIA are a rare exception to their aversion for partnerships - perhaps those last two categories are irrelevant.  

Of course the other question is whether any writer would have the chutzpah to write up London Calling over a month after it happened.  A very good question I suppose...

What Are We Supposed to Call PacBio's Short Read Chemistry?

PacBio presented new data and unveiled pictures of the instrument that will implement the short read Sequencing By Binding (SBB) that they acquired with Omniome.  But curiously PacBio hasn't unveiled what they are calling the platform, beyond the hardly pulse-raising "SBB".  

In their presentation, the system has amazingly high Q-scores out to 150 bases -- which makes many wonder how much farther they could read.  While so much can be successfully sequenced with paired ends, sometimes you just want longer reads to power through simple repeats.

They also weren't afraid to fire shots directly at competitors, with a slide purporting to show the best overall SNV and small indel calling of the short read platforms.  PacBio also showed SBB consistently sequencing perfectly 27 and 28 base runs of A/T as well as calling a one base insertion in the first element of a long VNTR array. Plus accurately calling a variant from a doped sample where the variant was present at 0.125% of the reads.

They're also running the system a lot -- over 1500 runs for this year! That should give a great database on reliability, though of course it makes a difference if that's 1500 runs of really boring libraries or a wide range of challenging samples.

Launch is still slated as first half of next year -- which of course can mean January 1 or 11:59 on June 30th.

What's Up With Illumina Chemistry X?

Illumina appears to either be very cagey or just not being aggressive about their Chemistry X, which was announced at JP Morgan.  X is supposed to be faster, more accurate and cheaper.  But which if any existing hardware will support it either straight up or with a simple upgrade?  That's a mystery that Illumina isn't revealing yet -- they've hinted that it will be supported on some platforms but which?  The newest NextSeqs and NovaSeq would be the obvious choices, but given the differences in optics between the two maybe only one will be supported? iSeq seems to never get any love -- or upgrades -- so that seems unlikely.

Nor was any information given about launch date.  Is X something Illumina tipped early to try to distract from the onslaught of new entrants, or are they still thinking they can garner big headlines with a big, well-timed announcement along the lines of how they stole Ion Torrent's initial thunder by launching MiSeq?  The longer Illumina waits to make their plans clear, the more customers will be tempted by the new entrants -- and at some point people must make their capital requests.

When Will Illumina Infinity Launch?

Illumina's other new initiative is the Infinity synthetic long read technology, which (along with PacBio HiFi and Oxford Nanopore) featured in Euan Ashley's keynote speech (which, strangely enough, actually opened the meeting as a keynote should). 

Ashley's talk gave some hint about data from the system -- read N50s were around 5 kilobases though there was a slide fixated on freak reads in the 20 kilobase range.  But not a lot about the laboratory workflow.  It does appear the DRAGEN hardware-accelerated software platform will be required for converting raw data into Infinity synthetic reads.

Illumina has been eager to show cases where Infinity can resolve what short reads can't, though they were torched on Twitter by PacBio before the meeting for one slide that seemed to show as many serious errors as successes.  And Illumina isn't being very forthcoming about how the system works -- most guesses are around PCR mutagenesis and perhaps Contiguity-Preserving Transposition. Nor is there any detail on launch date or pricing.

How Big an Issue is Ultima's Sequence Quality?

After decloaking in May, Ultima was also a major sponsor at AGBT and generally received favorable coverage at the meeting.  But the fly in ointment is a set of Tweets and a BioRxiv preprint from Lior Pachter and his student (just minted with a Ph.D., congrats!) Sina Booeshagi which looked at first one and then another Ultima dataset which is publicly available and put forth the case that problems with read accuracy significantly deflate Ultima's claimed price-per-base advantage (as in, if you claim to be 1/5 the cost but only 1/5 the reads are useful, then there's no actual savings).  Pachter and Booeshagi also found an example of a medically important gene TMSB4X with a hairy homopolymer in it -- and apparently variation in the homopolymer length is an important type of genetic variation.  Because of this homopolymer, a lesser number of reads mapped there, reducing comparability with Illumina data.  If nothing else, expect to find this gene in future PacBio SBB presentations!

They also point out that because Ultima must read all the way through certain single cell library inserts to get to critical UMI and/or barcode information, that data will be the lowest quality and many UMIs may not be decodable.

Ultima won't launch fully until next year, though they snagged a major endorsement and customer in the form of Exact Sciences.  Some observers have pointed out that Exact's cell free sequencing assays may be a great fit for Ultima, as they aren't trying to call specific mutations but rather call methylation (or copy number?) in bulk.  

The Pachter & Booeshagi piece illustrates the value of releasing data early -- there's now catnip for bioinformatics developers to attack the problem.  Indeed, they showed that their own pseudoalignment methods boosted sensitivity, though not in a huge way.  Perhaps other methods (homopolymer compressed mapping??) will be needed -- or perhaps the methods Ultima has already developed but unused by Stanford team do much better.  Conversely, Pachter & Booeshagi were able to legitimately grouse that most of the Ultima datasets are not available (and seriously, "available on request" is effectively unavailable) -- companies are smart to flood SRA with as many datasets as they claim to have (well, maybe all 1500 PacBio SBB from this year would be excessive).

How Disruptive Will BGI Be?

BGI has made it official: they'll be leaping into the U.S. market at end of summer with their CoolMPS technology.  As others noted, BGI has some really big systems that can also compete with NovaSeq or Ultima -- and which have the advantage over Ultima of having an extremely similar error profile to Illumina data.  

BGI also has plays in microfluidics-driven single cell sequencing and spatial sequencing, though some of these may be too early stage to be commercial.  Still, the possibility of BGI attacking multiple markets is very real and could make things troublesome for companies like 10X or the host of spatial players.  

But, BGI has the baggage of being closely tied to the People's Republic of China and the People's Liberation Army, and even if you don't object to that (I've already heard from one sequencing facility leader who does) the tense geopolitical situation between the U.S. and China will give many pause -- could some future trade embargo disrupt the ability to obtain consumables?

Will Tweeting AGBT Ever Be Big Again?

The number of people tweeting from AGBT seems to shrink every year -- though my deepest gratitude for the dozen or so regular contributors.  It is also notable that the sequencing vendors really seem to be slacking on their Twitter presence -- they should study Oxford Nanopore for their aggressive and informative Twitter coverage of their meetings.

And by the way, remember to title your documents! PacBio's presentation has a title starting with "Example integrated logo photo cover image; title automatically r..." (I can't figure out how to see the rest of this gripping intro).

Well, that's a wrap.  Time for my post-posting typo and fact checkers to do their welcome work.


Anonymous said...

There was a delightful series of questions, all rapid fire from one person after one of the Ultima talks, that was more or less:
"You mention the $100 genome, does this include sample prep costs?"
"Does this include instrument amortization?"
"Does this include any bioinformatic post-processing beyond simple fastq generation?"
"No further questions your honor."

Anonymous said...

In a conversation with our Illumina rep earlier this year they indicated to me that as far as what's been communicated with them thus far (not official, subject to change etc etc), the NovaSeq will be able to run Chemistry X.

Bertrand said...

Excellent and very informative, as usual, thanks for the effort!
Typo: There is a missing line at the beginning of the paragraph on Singular Genomics