“we found that the Illumina iSeq 100 used a very outdated implementation of BIOS firmware using CSM mode and without Secure Boot or standard firmware write protections”

A number of Illumina sequencers are based around x86 compute. In particular the MiSeq, MiniSeq and NextSeq 550. For practical purposes I view this as somewhat of a non-story. The iSeq 100 is a wildly unpopular sequencer, likely chosen because it’s relatively cheaply available on eBay.

And I wouldn’t personally view the issues here as particularly unique to DNA sequencers. A bigger issue is that these instruments are sometimes run without user authentication and connected to the open Internet.

What’s a little more interesting is that they provided some internal pictures of the iSeq:

Read more

Highlights:

• Illumina clinical lab directors get paid $330K USD in salary and RSUs worth $150K

• Illumina appear to have had some compliance issues that caused some concern.

Way back in 2022 Illumina fired one of their lab directors, this resulted a lawsuit.

The lawsuit alleges that Illumina fired a director of one of their clinical labs because they were discussing concerns about the facility being out of compliance with various statues and regulations:

“plaintiff attended a teleconference staff meeting chaired by X, Senior Director of Quality at Illumina/Verinata. Other staff member in attendance at this meeting were X. At the meeting, plaintiff expressed serious concerns about the multiple ways in which the laboratory facility (which his assigned duties and responsibilities required him to assure was in compliance with federal and state statutes and regulations) was out of compliance with those federal and state statutes and regulations. After he began to speak, X abruptly terminated the meeting and privately stated to plaintiff that he had “no right to inform her staff” that anything pertaining to the laboratory was out of compliance.”

The full details are a matter of public record (and available in the court filing linked below). These also give us a useful data point to calibrate salaries as the full employment conditions and contracts are included. This lab director received $330K USD in salary and RSUs worth $150K. That’s a bit higher than I was expecting, but there you go!

The lawsuit then gets more complicated… the arbitration documents suggest that the lab director was also running their own consulting company, and that they copied a large amount of confidential data from their Illumina laptop when they quit.

Seems like a horrible mess! Full court filing with all the details below:

Read more

Bitsof.bio is an overflow substack for ASeq, where I ocassionally publish supporting documents and opinions. It’s also cheaper and helps support all my writing. If you’re already a subscriber on ASeq and would like a complementary paid subscription here, ping me and I’ll add you!

Genapsys went into liquidation back in 2022. One of the fun things about liquidations/backrupcy is that creditors get made public. This gives us some insight into what tools/consultants were being used. Poking through the records pulled up a number of interesting vendors (listed below).

In particular three different companies appear to have been involved in the fabrication of their semiconductor DNA sequencing chips…

Read more

The Nucleus CEO reached out and said it was their investor that did the salute, not him (see his comments below). I’ve corrected the previous article to reflect this and apologize for my misinterpretation of the TechCrunch article.

TechCrunch confirmed this was Delian Asparouhov one of Nucleus’ investors and Founders Fund Partner, who was the other person being interviewed in the article.

In any case, I think this is a broader issue than one particular investor and my original article contains my views on Eugenics.

Nucleus CEO Comments

Based on this TechCrunch article, which ended with:

When asked if linking things like blue eyes, blonde hair, and IQ could be interpreted as eugenics, he clarified with a laugh, “I said brown hair!”

Then, miming the same hand motion that Elon Musk performed following President Trump’s inauguration, Asparouhov joked, “My heart goes out to you.”

You can decide for yourself if it was meant in jest of not. But in a discussion on eugenics, to make a gesture that has widely been interpreted as a Nazi salute while referencing the excuse given for that salute:

Is a Nazi Salute.

I think The Producers is a funny film! From what I remember the Great Dictator was pretty good too! Jokes about Nazi’s can be funny!

I don’t get the joke here.

What’s the joke?

Something about a group of people who would have killed my son, Eugenics, and Elon Musk… is really funny. So I’m looking forward to someone explaining the joke to me.

It reminds me of an event earlier in my career… Sitting down with a genomics executive in a pub… for some reason he decided to start explaining that he got into genomics for political reasons… to paraphrase:

“I want to eliminate stupid people. That includes all religious people, because all religious people are stupid.”

I guess it was a joke.

I don’t get the joke. Can someone explain the joke to me?

To be clear, eugenics is both morally wrong and scientifically flawed. It’s part of a movement that led to what is generally considered the worst atrocity of the 20th century. And yet I’ve seen this image casually flashed up on slides in scientific presentations without comment:

I don’t consider my views on this topic particularly extreme. Unlike some, who have categorized technologies like NIPT as a form of eugenics, I’m broadly in favor of them. But understand their is nuance there too, as this article states:

the rhetoric of choice is meaningless; to knowingly carry to term a baby with Down syndrome “cannot be a real option when society does not truly accept children with disabilities or provide assistance for their nurturance.”

And yet, I remain broadly in favor of NIPT and don’t believe it can be considered “considered a form of contemporary eugenics”. When decisions are ultimately made hopefully by individuals who retain their own reproductive rights.

What I’m not in favor of in minimizing the dangers of eugenics and making mocking Nazi salutes when addressing the topic.

Pseudo-scientific eugenics ideals can clearly can be used as a tool of political control. There are folks in positions of power who are quite happy to promote these ideas. Prominent scientists have promoted these ideas.

And I’ve seen them unashamedly professed in industry too.

When I was told that someone wants to use genomics:

“to eliminate stupid people. That includes all religious people, because all religious people are stupid.”

I didn’t say anything… I didn’t say anything to the religious person sitting next to him either… I don’t want to make excuses for that.

But perhaps I should be better prepared to say something if it happens again…

Because it feels like it might happen again.

I decided to cut a small aperture in some black tape and use this as the baseline for comparison. This meant I could then cover up the flowcell regions outside of the channel itself so I wouldn’t get any contribution from the black material on the NextSeq 550 flowcell or the bonding material used on the Miseq flowcell. The original images are shown below. The figure above roughly aligns everything for comparison.

Based on these measurements the MiSeq flowcell still looks like regular borosilicate slide glass to me. The NovaSeq is transparent much further into the UV range suggesting to me it’s closer to (or is) fused silica. But weirdly the NextSeq 550 glass is somewhere in between.

So… I still don’t know what this is! But at least we have some slightly better data…

In part this has cost implications (borosilicate is cheaper than fused silica). But I’m also just curious to know how flowcell development has progressed…

Insert size is probably limited by a number of factors. One concern on patterned flowcells is that templates might form bridges between wells resulting in duplicate reads. With super-resolution, and well spacing going below 300nm this could be a factor.

However this doesn’t mean it’s the only factor. The original Isothermal patents were not on patterned flowcells. But also show a fragment length dependance on amplification:

In the figure above T90 is 90bp, T605 is 605bp etc. The Y-axis is normalized for template length and therefore gives a fluorescence based estimate of the cluster size.

At very short template length efficiency drops (they cite the notion that short templates are rigid and can’t bend as easily).

But beyond that as template length increases efficiency drops. I didn’t see a justification for this in the patent. I assume longer fragments are more likely to form secondary structure etc.

Illumina run a fixed cluster generation protocol. As such it seems possible that they might be able to grow clusters from longer fragments under different conditions. However as longer fragments are likely to form duplicates anyway (on patterned flowcells), they likely don’t want to support this. But perhaps a hacked recipe might allow longer insert sizes? (at the risk of duplicates).

gdub42 also dug out this really interesting paper which shows fragment length dependance on read quality (particularly apparent in the second read):

Would be really fun to hack around with recipes and how growing clusters longer/for more cycles might effect this…

The performance characteristics of CLR and CLR differ in some respects. The reads generated by CLR naturally have a relatively high error rate. However, because these errors are largely non-systematic a CLR consensus can correct for these errors (a similar process exists in CLR). Of course company statements suggest that CLR is “not a synthetic long read technology”. This is debatable, but in contrast to this CLR clearly is a “single molecule level” technology. In my opinion this gives CLR a slight edge over CLR.

In the case of CLR, reads can benefit from correction using non-CLR reads from the vendors short read platform. Similarly CLR reads could also benefit from correction using non-CLR reads from the vendors short read platform. Or vice-versa! There’s nothing to stop you doing it the other way round! But there are likely cost advantages to going with the CLR vendors short read platform (this would also be true in CLRs case).

A full systematic evaluation of CLR over CLR would be interesting, particularly if assembling the longest “not synthetic” read is your goal. My guess is that CLR might have a slight edge over CLR here.

To be honest I don’t get the impression that CLR has played well in the market so far (this is of course true for CLR).

Overall I’d suggest that other technologies such as PacBio Hifi reads (where you need high accuracy) or ONT reads (where lower accuracy is acceptable, but you need the longest reads possible). Might be a better fit for most applications.

In any case I will keep watching for development in CLR (and CLR). CLR always seemed like a neat approach but perhaps one limited to niche applications. But, I hope we get to revisit CLR (and CLR) at some point and see developments based on these approaches.

What is the BGI?

This is a seemingly simple question to which I have no simple answer. I mean it’s a research institute right? That’s in the name? Well… kind of… but not really.

I’ve written about my confusion with the BGI before, but let’s try and summarize. The BGI was founded in 1999 with a mission to contribute 1% of the data to the human genome project.

After the completion of the human genome project the BGI relocated to Hangzhou in exchange for local government funding, and then in 2007 they announced that they were to relocate to Shenzhen to establish China’s first private non-profit research institution.

So from what I can tell they started off firmly as a non-profit research institution and remained so until at least 2008.

2010 seems to have been the turning point for the BGI. They received a 1.58B USD line of credit from the China development bank, and funds from Shenzhen Capital Group. It also looks like Taikang Life Insurance and Sequoia may be an investors2.

Then in 2013 after a long and slightly painful process they acquired US DNA sequencing startup Complete Genomics.

Originally the BGI bought DNA sequencers3 from a US company, Illumina. In fact, they were probably their largest customer. Purchasing 128 HiSeq 2000s DNA sequencers which had a list price of ~$700K in 2010.

Then in 2015 they started making their own DNA sequencers! And finally on the 14th July 2017 they were listed on the Shenzhen stock exchange completing their transition in to a somewhat surprisingly highly commercial research institution.

So… BGI started as Illumina’s biggest customer and ended as their biggest competitor! (well… kind of4)

Then they all lived happily ever sued each other! This was more or less resolved in 2022 when they agreed not to sue each other for at least a little while.

What Is Happening With This Bill?

As far as I can tell the bill would make it very difficult to spend US public funds on BGI (and subsidiaries MGI and Complete Genomics) instruments.

As the bill says:

It is the sense of Congress that the time has come to stop United States taxpayer dollars from flowing to foreign adversary biotech companies like BGI that have ties to the PLA; and prevent United States taxpayers from buying biotech equipment from foreign adversaries that facilitate the transfer of United States persons genetic data to a foreign adversary.

In light of the BGI being a competitor to Illumina (a US company) you might be forgiven for thinking this is a protectionist measure. Well… maybe… but the thing is the BGI isn’t really much of a threat to Illumina.

While I described BGI as “their biggest competitor” above. The reality is that Illumina doesn’t actually have any very big competitors and maintains something like an 80%+ market share.

Their biggest threat isn’t really the BGI, but a new crop of smaller startups. BGI machines have been around for years and don’t seem to have gained much adoption. In part, because they don’t seem to be massively cheaper than Illumina instruments…

Genetic Weapons

I’ve seen news of the bill quoted along side think-tank reports on genetic weaponry discussing China’s exposure in US Genetic testing:

The report goes on to describe how they believe that China is nominally engaged in”biological activities with potential dual-use applications”. And cite a number of sources where PLA officers have “suggested that China should consider biological weapons” e.g.

As noted previously, in 2015 a PLA Daily article co-authored by then-president of the PLA’s Academy of Military Medical Sciences He Fuchu (now AMS Vice President overseeing implementation the PLA’s civil-military fusion strategy) entitled “Biotechnology will become a new strategic commanding point for the future military revolution” argued that “the weaponization of biological bodies will become a reality in the future, non-traditional combat styles will take the stage, ‘biological frontier’ will become a new high frontier of national defense.

The report is interesting, and suggests that the “ethnic diversity of the US population provides a protective effect that largely undermines the potential strategic payoff of a genetic weapon system”.

But overall, it didn’t seem like this report was making a very strong case for blocking Chinese access to US genomic data.

Surveillance

The bill itself seems more focused on surveillance applications citing a National Counterintelligence and Security Center advisory from 2021. This report discuss China’s use of genetic surveillance domestically. And suggest that the PRC could use this information for surveillance, manipulation, or extortion:

For instance, vulnerabilities in specific individuals revealed by genomic data or health records could be used to help target these individuals. Data associated with an embarrassing addiction or mental illness could be leveraged for blackmail. Combine this information with stolen credit data indicating bankruptcy or major debt and the tools for exerting leverage increase. Such data sets could help the PRC not only recruit individuals abroad, but also act against foreign dissidents.

The advisory is from early 2021 (the middle of the pandemic) and mostly seems to focus on issues around COVID-19 test kits. So it seems a bit weird that it’s resurfacing now.

But it makes a bit more sense to me than the genetic weaponry angle.

Summary

I’ll likely never understand the political motivations behind this bill. And I don’t have much insight into whether it’s likely to pass or not…

However, even if it does I’m not sure how much of a difference to the market it will make. The one US customer I’m aware of (Nebula Genomics) is a commercial service, so wouldn’t be covered by restricting use of public funds.

And I don’t get the impression much public research5 uses BGI instruments anyway…

More significant is the chilling effect this will likely have on the purchase of Chinese DNA sequencing instrumentation and tests in general. And the bill simply being proposed is probably enough to discourage many users already.

I decided to have a quick play with the NextSeq 550 XY Stage. This is a custom design interfaced via USB based around an STM32F103. In order to get the board to power up you need to supply it with ~5V via J3 as shown in the diagram above. A USB serial device is then exposed.

The stage uses an encoder wheel from US Digital. A full encoder assembly from US digital costs ~$100. So the whole XY assembly is unlikely to have a BOM cost of more than $500 in my view, call it $1000 if you like. NextSeq 550 sells for >$200K so this isn’t really a significant cost.

Would be fun it figure out the USB protocol. But I suspect more effort than it’s worth for a quick side project.

It’s been tough finding much information in patents, but I think I have a somewhat reasonable idea of what’s in this box now. So, here are a few brief notes.

A few Illumina presentations show parts of the NextSeq 2000 optical bench at low resolution:

These look somewhat similar to those shown in patents for structured illumination super-resolution systems (which it appears Illumina use in their current instruments):

Based on these patents (see below). The camera seems to be something in the IMXn83 series, which likely cost a few hundred dollars. Probably about the same COGS as the MiSeq cameras.

The patents appear to suggest that lasers are used rather than LEDs (like the MiSeq). And Illumina posts on SIM suggest the use of lasers.

I suspect these lasers and the associated optics are the most expensive parts of the NextSeq 2000. The lasers used in the Genome Analyzer cost something like $10K, so a rough estimate of $20K doesn’t seem unreasonable.

The fluidic system on the NextSeq is also a departure from previous instruments. It looks more like the iSeq fluidic system. The flowcell interfaces directly with the reagent cartridge and from what I can tell there is no fluidic coupling between the cartridge and the instrument.

I’m a big fan for this, and it must make for a much more robust platform.

Overall I’d kind of expect that the COGS of the NextSeq 2000 is ~$20K to ~$30K higher than the MiSeq.

Sensor Notes

Patents describe a system using a 5472×3694 pixel sensor, and 10.63x magnification. Field of view covers 814 microns. This suggests a pixel size of between 2 and 3 microns. Googling all the resolution and pixel size the best matches I can find are the IMX183, IMX283 or IMX383 sensors. These are all relatively cheap sensors which retail <$1000 in single units.

Quotes below from this patent.

“The field of view 4100 can be defined relative to a sensor area 4102. In some implementations, the sensor area can include about 5472×3694 pixels. In some implementations, the field of view 4100 includes multiple tiles. For example, the field of view 4100 can include a 7×7 tile region. For example, one tile can include about 512×512 pixels (about 116×116 μm). In some implementations, the field of view 4100 can have a width of about 814 μm (about 3600 pixels). In some implementations, the field of view 4100 can have a height of about 814 μm (about 3600 pixels).”

Cameras with near specified resolution:

The amateur astronomy folks seem to like the IMX183/283/383 too, which is always a good sign (suggests good low light performance/QE/readout noise).

“It's all rubbish isn’t it... I mean…Transplants, antibodies, we manufacture genes, we can produce birth ectogenetically, we can practically clone people like carrots, and half the kids in this ghetto haven't even been inoculated for polio! We have established the most enormous medical entity ever conceived, and people are sicker than ever! We cure nothing! We heal nothing! The whole goddam wretched world is strangulating in front of our eyes! That's what I mean when I say impotent! You don't know what the hell I'm talking about, do you?”

The Hospital was written by Paddy Chayefsky and developed out this own experiences dealing with New York hospitals in the 1970s. But in some ways the monologue above feels like it could have been written yesterday.

The film is well worth checking out, and you can find whole monologue expertly delivered by George C Scott on YouTube.

In this short series of posts about conversations with clinicians. I’m going to continue to look at why some doctors don’t seem to be particularly fond of diagnostic tests.

Here’s another fragment of a conversion I had with a Neurologist at Chiba University Hospital:

Me: Fragile X is not tested for in Japan, but it is common in other countries. please explain.

Doctor: Well, if the ear has a major abnormality, you can do it.

To give some background. My son has profound intellectual disability1. Here in Japan he wasn’t offered Fragile X testing. Elsewhere this is a standard test, as the American Society of Pediatrics write “Fragile X genetic testing should be performed in all.” (cases of intellectual disability).

Fragile X often results in other symptoms along side intellectual disability such as a “long and narrow face and large ears”. Which is perhaps why this doctor is suggesting testing only when the child has abnormal ears.

Why then is the American Society of Pediatrics in favor of testing in all cases?

Informally the American Society of Pediatrics writes about this on their blog citing one case in particular:

“His parents came into my clinic when Brian’s developmental pediatrician referred him for a genetic evaluation. Part of the genetic work-up we did was a Fragile X DNA test. When those results came back positive, his parents were surprised to hear that Brian had Fragile X syndrome.  They thought that he was just a little slow with his language and that he might be autistic. They didn’t think he looked different enough to have a syndrome.

Not all children with Fragile X have distinctive features, which is why it’s important to test.”

Let’s try and put some numbers around this. One study showed that <50% of individuals have any one facial feature associated with fragile X:

Another study showed better results, in this case 72.2% of boys showed increased ear length, 83.3% an elongated face:

In both cases it’s relatively clear that if you look at physical features alone there will be a high false negative rate. Results from the Netherlands suggest that >50% of fragile X cases are undiagnosed2. These and other results have led many to consider facial features unreliable in diagnosis:

“Physical features involving the craniofacies (long face, prominent forehead, large ears, and prominent jaw) not readily recognizable in the preschool-age child become more obvious with age. Only a subset of affected individuals have typical physical features of FXS and presence of these is not reliable for diagnosis.”

And is perhaps best summed but by these authors: “Despite many attempts to classify the facial features in fragile-X subjects, conflicting opinions remain on the nature of the anomalies and their degree of severity.”.

So Why Don’t They Test?

I don’t know. So I can only speculate. I assume if you pre-screen based on physical features a much higher percentage of tests come back positive. The literature appears to back this up3.

If you don’t pre-screen a much lower fraction of tests are going to come back positive. This is expected. The statistics just work this way and you’re either going to pay with a false negative rate or by running a large number of additional tests. The tests being accurate, low cost and relatively non-invasive, many would consider it worth while to test in the case of intellectual disability.

I suspect, as previously mentioned, many clinicians don’t like running tests which have a high probability of a negative result. And in any case don’t see an accurate diagnosis as valuable if there are no clear therapeutic opportunities4.

Oh No! You’re not subscribed! Perhaps you’d like to consider subscribing!

On Monday we went into Chiba University Hospital to see the Neurologist for my son who is intellectually disabled.

Part of our conversation went like this:

Me: What are the standard tests that he should have done.

Doctor: Um... Basically, in Japan, rather than thoroughly investigating the cause, the standard approach is to follow along and make it easier for this child to live.

Me: So you don’t do any tests?

Doctor: If something different that can be done...

Me: I understand there are no therapeutics for intellectual disability1. If he has fragile X it effects my daughters children. So of course you do tests.

The basic standard as outlined by the American Academy for Pediatrics is fragile X, Microarray (CMA) and metabolic testing before moving on to brain imaging, gene panels, whole genome sequencing etc.

Their 2004 report elucidates the reasons for testing, even in the absence of therapeutics quite succinctly:

“Accurate etiologic determination, despite the fact that many disorders have no specific therapeutic interventions, has specific implications regarding treatment, prognosis, ongoing medical management of associated conditions, assessment of recurrence risk, counseling of families if there is a risk of recurrence, and implementation of prevention programs. Determining causality also empowers the affected family in planning for their child and limits further unnecessary testing.”

In the case of intellectual disability, there are a number of potential genetic causes (fragile X among others) which would be a consideration in family planning, both for parents and any other siblings. In these cases, even knowing what “it isn’t” is of value.

From my personal standpoint as a parent doing everything possible to understand my sons condition feels like it has value in itself.

Some might say a bias against testing is peculiar to Japan. But in my opinion there’s enough evidence that it’s relatively pervasive… particularly outside of research driven institutions.

More Common Conditions

Back during COVID I tried pitching better tools for infectious disease testing. The idea being that if we could develop something better than qPCR we could test individuals for all known pathogens and get an early warning on new unknown ones2.

If we could make this the standard of care for even some fraction of patients showing “Flu-like symptoms” we would as a bonus get an early indication of a new virus in circulation. Potentially helping us avoid pandemics in the future… or at least giving us a head start of vaccine development.

The problem here is that clinicians need to want to know what infections their patients have. And there has to be an expectation from patients that if you visit the doctor with “a bad cold” they won’t just give anti-inflammatories/fever reducers and tell you to get some rest. Rather they’ll do a test and tell you want is wrong with you.

That would require a huge cultural shift.

The Ideal

My ideal is that more information3 is always better. Understanding puts us in a better place to move forward both on an individual and societal level4.

So I suppose my ideal would be that we would start with this goal in mind. That5 every tool at our disposal is used to determine the cause of a condition to… fully elucidate the etiology.

This applies as much to intellectual disability as it does to the common cold. First obtain what information we can, then decide how to move forward6. But where the information in and of itself has value7.

Where that information, as well as applying on an individual level, can be used to guide social policy, therapeutic development and basic research.

And The Reality

Without this information being valued, the reality is that it remains risky to develop new diagnostic tools. You end up in a Catch-22 situation, without new diagnostic tools you may not be able to identify new potential targets for therapeutic intervention. But without therapeutics clinicians won’t run the diagnostic anyway.

That makes the development of new diagnostic tools risky unless you’re developing something that’s just “slightly more accurate”, “slightly faster” or “slightly cheaper”. A broader or new diagnostic is unlikely to gain adoption in anything but small niches.

And as such many VCs will be wary to fund much tools.

Falls The Shadow

I suspect everyone is trying to do their best for patients and to improve society in general. But as I commented to one VC recently: “This feels like something where it would be of clear societal benefit. But it’s difficult to see how we map a course from where we are to broad adoption”.

For some reason I’m reminded of a popular quote from Shaw:

“The reasonable man adapts himself to the world: the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man.”8

When I request diagnostics for my son, or suggest routine meta-genomic sequencing for infections I feel I am often perceived as the "unreasonable person”. Unfortunately I suspect I have little choice but to continue to move forward with my “unreasonable” requests.

Oh No! You’re not subscribed! Perhaps you’d like to consider subscribing!

The first cases of COVID-19 are suspected to have occurred in late November 2019… which is why I guess when ProMed started reporting “undiagnosed pneumonia” a few days ago everyone got a bit… excited:

A more recent comment on ProMed suggests reports are consistent with Mycoplasma, Why exactly we don’t know at this point if this is even a bacterial of viral infection in a mystery to me…

In any case “it’s probably going to be ok”… but it’s an interesting chance to review how we've progressed since 2019 in diagnostics and surveillance… what do we have that would help us through a new pandemic?

Honestly the answer is “pretty much nothing”. But let’s review!

At-Home Molecular Tests

These things are pretty cool. It’s like having a little qPCR machine that you can use once to see if you’re infected and throw away. The engineering and cost optimization that went into devices like the Lucira is impressive.

Unfortunately most of these seem to have been commercial failures and made a limited impact on the pandemic.

People are just unwilling to pay $50+ for a single use test, even if it is more accurate than lateral flow…

These products were already in development prior to COVID-19 but the pandemic likely helped push them to market.

Surveillance Infrastructure

We’ve build and expanded a bunch of (sometime questionable) infrastructure for COVID-19 variant tracking. There are also generally accepted methods and protocols for COVID-19 variant tracking. Getting this up and running and communicating the importance of variant tracking in a new pandemic would likely be far easier next time round.

While a lot of the labs doing variant tracking disappeared or refocused, I suspect we could spin them back up quickly.

Environmental Surveillance

We have things like the national wastewater surveillance system. And we’ve talked a lot about meta-genomic waste water surveillance. We’ve talked about it, and we’re moving towards it. And we think it could be a really good idea…

Companies also think you should spend money on it.

But I’m as yet unaware of anyone actually doing it in a coordinated way, with significant funding, on a global level.

But perhaps we could shift our COVID-19 wastewater monitor systems over to the next pandemic.

It’s Embarrassing And We Should Feel Embarrassed

To be frank, none of the above puts us significantly ahead of where we were in November 2019.

I suspect our oligo synthesis pipelines might have got a bit better, we can probably deploy qPCR kits a bit faster than the 4 months it took to ship 1M tests in the US for COVID-19.

There are have been no advances in sequencing from which a future pandemic would benefit. The NextSeq 2000 was probably the most cost effective instrument for variant tracking… it probably still is.

We have no new tools which will give us an early warning of an impending pandemic or potentially help us contain its spread.

It’s kind of embarrassing! And as someone who works in this area, I’m particularly embarrassed!

We could have cheap, effective tools for identifying new pathogens… helping bootstrap vaccine development and contain spread.

But somehow humanity has decided that it’d rather react to crises than create safeguards against them.

• Q versus emQ: Which shows how well calibrated Q scores are.

• Q yield plots: How big is the population of each “Q”.

• Overall error statistics.

Which Samples And References?

The first issue was figuring out which datasets to use. Turns out HG002 which is part of the Genome-in-a-bottle sample set is the best studied human genome we have. It fast seems to becoming the a standard sample to run when validating a new sequencing platform.

New human genome references have been developed for this sample. In fact, two references one for each of paternal and maternal chromosomes. The best reference to use seems to be just concatenating both of these individual reference genomes (as evaluated here).

With that process established we actually have to do the alignments. I used Dorado aligner (which uses minimap2) do align everything. This isn’t ideal. I suspect just using stock minimap2 would be fairer and will likely work on short read data as well as long.

Which Datasets?

Next thing is to look for datasets. You might think this is as easy as grabbing them from whatever source and then aligning them… it’s not unfortunately.

Many of the datasets I found had issues… I found that in general it was best to convert whatever you’re given to a fastq and then use that for the alignment input.

Dorado aligner (and I assume Minimap2) will work from BAMs (probably crams too?). Theoretically you could use these… However I found that the resulting BAMs would contain references to any original alignment1, this would cause BEST to fall over (correctly) when it could find the referenced location in the reference.

Going back to fastq naturally strips out any alignment information and gives a clean input to work from2.

With a clean input you can align and run the output through BEST.

I’ve so far used BEST to perform this for the following datasets:

• An Oxford Nanopore dataset called with the “fast” base caller model.

• An Oxford Nanopore dataset called with the “sup” model and duplex model.

• A Pacific Biosciences Revio dataset.

• An Illumina HiSeq X Dataset.

• An Ultima Genomics Dataset.

• A Qitan Technology Dataset.

These runs have all be fairly instructive! I hope that vendors start running their benchmark datasets though BEST and providing benchmarks based on this.

While these statistics don’t provide a complete picture of the performance of a particular sequencing platform, they do provide a valuable insight into the broad performance characteristics.

Perhaps a larger project where we look at how accuracy has evolved over time would be fun? Thoughts on future projects welcome on the Discord!

In that post I moaned (needlessly and in a thinly veiled attempt to gain subscribers) about the following statement:

“Illumina invented a profound new measurement technology”

My claim was that Illumina didn’t invent anything (profound and new), they bought it.

The objections mostly feel like a misunderstanding… as my intension was not to say Illumina haven’t contributed anything to the development of the next-gen sequencing. Just that the basic approach came from Solexa and the core of that approach remains.

What Did Solexa Contribute?

I’ve written about Solexa a couple of times before. But let’s briefly review.

There are 3 main components1 to the Solexa approach:

• Sequencing-by-Synthesis

• Reversible Terminators

• Bridge Amplification

Bridge amplification came from Manteia Predictive Medicine and honestly, was probably a fundamental contribution which is all too often overlooked. So I should be clear, I’m not saying Solexa invented everything either… just that Illumina didn't.

Solexa was originally a single molecule sequencing company. Their first instrument (the Genome Analyzer) was designed as a prism-style TIRF platform which I imagine could have supported single molecule imaging.

It wasn’t a great instrument by any stretch of the imagination. More or less a lab rig thrown in a box. But it shipped. And post-Illumina acquisition it shipped largely unchanged in reasonably large numbers.

When I was at Sanger, we had ~20 of them. Other large genome centers also started buying them in large numbers. It kind of became clear that if you didn’t you’d be left behind.

There was nothing else on the market that was competitive. None of SOLiD, 454, Ion Torrent, Helicos, Complete Genomics or PacBio2 had the necessary throughput or data quality.

Illumina Executed Well

Post acquisition Illumina made a number of strong contributions to the Solexa approach with the development of the HiSeq and subsequent instruments. These include:

• TDI Imaging (scanning across the surface using a linear-ish sensor).

• 2 Color chemistry.

• Patterned Flowcells and Ex-amp.

• Super-resolution Imaging.

These have all been important in enabling cost reduction, throughput increases3 and improved data quality.

They don’t in my view alter the basic measurement approach.

Illumina Without Solexa

Where would Illumina be if they hadn’t acquired Solexa?

Illumina was only just turning a profit4 at the time of the Solexa acquisition.

Assuming Illumina had remained interested in NGS who else could have they acquired? 454? Ion Torrent? Perhaps they would have acquired Helicos and developed an alternative surface amplification approach?

Thing is… Illumina have been pretty terrible about releasing fundamentally new approaches and have mostly grown through acquisition. Illumina own the MspA nanopore IP, but have failed to execute on it. So it’s not clear to me that a new sequencing technology would have been developed in house.

We’ll never know what would have happened. But I personally I doubt we’d we talking about Illumina today if they hadn’t acquired Solexa.

Solexa Without Illumina

Let’s face it British companies often fail to play well internationally. I suspect the Sanger Institute would still have bought those 20 Genome Analyzers. I suspect other Genome Centers would also have acquired instruments…

But would Solexa have released a HiSeq? Or NovaSeq?

This is less clear to me. Overall… it feels like Illumina was a good home for the Solexa technology. Illumina executed well. Other acquirers5 may not have. And Solexa might have had a hard time executing on the platform alone.

Who Cares Anyway?

I guess the question is who cares anyway? I mean this is all ancient history isn’t it?

Well… kind of.

The point is that some people are still somewhat bitter about the Solexa acquisition and wish Solexa had remained a “UK” company and believe that the $600M exit they got was too low.

Many folks from Solexa went on to Oxford Nanopore who put in place anti-takeover shares that prevent acquisition. You don’t need to take my word on the motivation:

“Medisense was an Oxford spinout that was sold to Abbott in 1996 for $876m and Solexa was sold to Illumina ten years later for $600m. The Solexa deal enabled Illumina, now worth $70bn, to become the world’s dominant sequencing company. “Myself and our CTO Clive Brown [a former Solexa employee] don’t see these as great successes. We saw this as a sad ending of what could be a globally dominant tech, and no revenues really have come back to this country,” - Gordon Sanghera quoted in the FT

Those anti-takeover shares appear to expire next year. I guess only time will tell how this plays out for Oxford Nanopore.

Twinstrand/Guardant

Twinstrand won a lawsuit against Guardant! I previously wrote about this, and speculated that Twinstrand’s formation might have in part been motivated/supported by Guardant’s success using what seems like a very similar approach. TwinStrand won damages of $83.4M… there will be appeals of course. But long term will Guardant be forced to license the technology from TwinStrand? TwinStrand laid off 50% of their workforce not long ago. Guardant also laid off 7% recently. But this seems like it should be good news for TwinStrand perhaps they’ll re-focus on technology licensing… or will be acquired.

Singular Call/Lay offs

The Singular Genomics slow motion train wreck continues. The full transcript of their earnings call is available here.

They shipped 5 instruments in Q3. Sounds like one of them went to the Broad. But this is pretty insignificant compared to Element who just installed their 100th instrument (Singular have shipped 16 total). One customer is using a reasonable amount of sequencing however:

“In fact, one of our earliest customer placements has ramped their utilization over the last nine months to an expected annualized pull-through rate of approximately $170,000.”

They also laid off 10% of their staff. They believe these actions will extend our cash runway into early 2026. And they still have a decent amount of cash in the bank: Net loss for the third quarter of 2023 was $22.4 million… Ending cash, cash equivalents and short-term investments, excluding restricted cash, totaled $190.7 million.

Read more

I recently wrote about the MiSeq cameras on the other substack. I also decided to take a quick look at the control electronics in the Miseq. But I don’t have quite as much to say about this, so you get it here on the second substack!

Paid subscriptions here (and the main substack) will help pay for broken bits of sequencers and are appreciated (I set the subscription here to the lowest allowable value).

As shown in the image above, most of the MiSeq electronics (aside from the cameras and a few other bits like the RFID reader) sits in the same box and the compute.

It’s connected to the main compute board via a long ribbon cable. This appears to be a serial interface (based on the PCB silk screen).

The control electronics sit on a couple of PCBs in mezzanine configuration. The top PCB looks like this:

The two identical layouts to the top are the LED drivers (Red and Green). Below this we have connectors that head out to various sensors around the device and two connectors for driving TECs (Peltier devices).

The TECs appear to have some pretty beefy associated power electronics… All this is fed from the same ATX power supply that drives the motherboard.

If we remove this we find the lower control PCB:

The top side for this is pretty bare. It have ATX power supply connectors, which it loops through to the compute motherboard. There’s a stepper motor driver (which is probably for one of the stages):

There would normally also be a PI piezo driver attached to this board I believe. This would be for the objective focusing module (at some point I will likely to a post on this so you know… subscribe)

Flipping this board over shows some more interesting components:

There’s old our friend the Spartan 6 (LX45)! To be honest, I doubt this is doing anything that probably couldn’t be handled by a DSP or microcontroller. As previously mentioned there are a couple of FPGAs in the cameras already. So the Spartan 6 here doesn’t have to do much heavy lifting.

That’s about it! I can’t see that Illumina have made much of an attempt to cost optimize here. The only consideration in that direction is perhaps the use of a off the shelf PC power supply. But I suspect that was just convenient (given they’re using it to drive a PC motherboard among other things).

The whole thing could be put together much more cheaply if you really wanted to. But when the instrument costs $100K there’s no real motivation to shave a few hundred dollars off the BOM cost!

That’s it for now! There should be some more teardowns coming soon! Including something quite… big… stay tuned!

I wanted to compare previous datasets to some Illumina data. So grab a random HG002 fastq from the PanGenome project.

The results are largely unsurprising. But it was interesting to look at some HiSeq data and see all those lovely (reasonably well calibrated) Q scores:

Here are the yield plots. BEST assigns a emQ of 75 to perfect reads, which here is the bulk of the data:

Overall the error rate seems very low, slightly lower than the previous PacBio dataset:

The identity here is actually 0.9988. So 99.88% accurate. 0.12% error rate, somewhere around Q29.

More or less what I would except to see!

Also… Century of Biology has like… 100x the subscribers I have here… so please subscribe to bitsof.bio!

Datasets here:

HG002_HiSeq30x_subsampled_R1.fastq.gz

https://s3-us-west-2.amazonaws.com/human-pangenomics/index.html?prefix=NHGRI_UCSC_panel/HG002/hpp_HG002_NA24385_son_v1/ILMN/downsampled/

The notes here describes this as:

Whole-genome data, downsampled to ~30x PCR-free Illumina 150bp (HG002, HG003, and HG004) to match the expected production data available for the 1000 genome samples (produced at the NYGC: ~30x 150 bp PE Illumina reads from the 2504 1KG samples (using ~450bp fragment size))

I aligned only the first read and used Dorado aligner (Minimap) which is consistent with previous runs.