专家称低成本板式单细胞技术对10x Genomics构成中长期竞争压力

[[00:01:58]]
How do lower per-sample pricing dynamics of plate-based combinational indexing platforms such as Parse Biosciences introduce the direct commercial revenue risk to 10x Genomics?

[[00:02:17]]
I think, I mean, it is a direct risk for 10x droplet-based technology. It's more like a mid-term or long-term risk, not the immediate one because still the 10x Genomics Chromium technology is the gold standard. It's the first choice for multiple applications. As we are introducing… Moving in the direction of utilising AI for target discovery, like a larger scale data generations, I think that's where the challenge starts to show itself. I think still the go-to technology is 10x Chromium for multiple reasons, but it's a direct risk and that it needs addressing basically.

[[00:03:20]]
Why do customer switching friction and trust patterns tend to favour alternative combinational methods over premium microfluidic droplet workflows?

[[00:03:33]]
I think, I mean, it's certainly the cost that's the major driver. It's a trade-off between sensitivity and specificity that 10x Chromium offers, but also like a community-wide acceptance, like when you go for writing publications, peer review publications or like using a data for very critical decision-making, still 10x Chromium technology is more trusted mostly because we have a lot more data with 10x Chromium. That's changing. Now with the plate-based technologies, they are also… Mostly by more data generations and being more trusted by the community through peer-reviewed publications through wider adaptation, now the dynamic is changing because now we are seeing that there is a real alternative path for the most critical challenge that we have so far, like is the number of samples that we can analyse with a single cell or the depth in terms of the number of cells that we can sequence. That was always… For any study that we design, the major bottleneck is the cost of the single cell. If that step is sort of like there's a way around that, I think that's something that we want to explore and we want to go after.

The process is slow, mostly based on how much do we trust or how much the community trust and how much benchmarking, do we need to get to the point that we can completely replace. It is something that is a question that comes up in every study design.

[[00:05:59]]
How successfully did massive cell perturbation repositories such as the Parse Bio 100 million cell dataset ultimately erode 10x Genomics' long-term positioning, which tends to be seen as the gold standard?

[[00:06:17]]
I think that's in general, for the community especially when it comes to target discovery and the use of AI becomes a critical element, especially now like tech companies that are leading AI space are entering the life science and drug discovery and they're initiating a team within their companies, the need for high throughput, larger scale data becomes very clear. That's sort of like what also the pharmaceutical companies that's very well-known. So far, it has been the limitation of like we can't really go beyond one million cells, economies doesn't support that, there's not enough funding. Plate-based technologies, especially like a few open source large dataset that they release like Tahoe-100M or 10 million drug perturbation experiment and the PBMC that they released and made it open source that the community started to adopt those. It became like a successful story that now either pharmaceutical companies expand on it by generating new data, expanding to new drug perturbations experiments or it does also make the CRISPR screening more economically reasonable in terms of funding. Over time, I think the community starts to trust, there are peer-review publications, and it's easier to make a case to go build plate-based technologies, especially for those larger-scale experiments that would economically be not feasible with the 10x Chromium.

[[00:08:47]]
How does the combination of intense competitive pricing pressure and plate-based barcoding flexibility ultimately drive labs to bypass microfluidic hardware upgrades entirely?

[[00:09:06]]
Plate-Based technologies are very easy to adapt. They do not require any specialised instruments, and you can adapt it pretty much with the standard instruments that are available in the lab and the process is not complicated and also given the current environment, it's very challenging to… Especially work labs that are starting that they do not have the 10x Chromium instruments in-house, it's very challenging to get funding to acquire new instruments, so like that large capital investment before any data generation, I think that really creates the opportunity for labs with financial challenges to adapt it because that's just a lower cost per sample. It's less than USD 100 per sample, we can run on plates. In terms of adaptations, I think that's also the major driver, the lower cost is driving the adaptation. I think it's the only main reasons for slower adaptation is really like the… I mean, 10x being gold standard in the community and having that customer base, and like a slightly better sensitivity. It's a very easy process to adopt the plate-based technology.

[[00:10:58]]
By how much does Scale Biosciences' technology ultimately lower raw sample processing costs? Does this reduction allow 10x Genomics to realistically secure high-volume screening contracts that were previously seen as economically unviable?

[[00:11:15]]
Yes. The entire process might be a little bit more complex to put a number on it. I mean, in terms of like the processing of the raw samples, I think 70-80% with 10x Chromium, it might be around like 400-500 per sample while you go down to USD 80-90 per sample in like plate-based technologies. I don't exactly know the Scale, but in general, that plate-based technologies is 70-80%. I think the major cost now shifts into labour and also sequencing cost still remains the same problem. Even like I think the other challenges as we lower the cost of the processing, now the idea is that we can have more samples, but that adds more cost from labour and from the sequencing perspective. The actual economy depends on more than just a raw sample processing. In general, since it makes the 10x compatible or comparable in terms of cost with the Parse technology, then because we see more adaptation of the Parse technology for larger-scale screening, I think the same will happen with the Scale because the cost-wise is similar.

In general, there is more trust than 10x when it comes to single cell because the community has accepted more, there is more customer base. After Qiagen bought Parse, I think that also in terms of reliability that improves. In general, I think it makes it compatible or competitive with Parse technology.

[[00:13:48]]
When a pharma account is designing large-scale genetic validation screens, how does the new plate-based baseline price point compare to traditional bulk sequencing economics?

[[00:14:08]]
I think it's going to be a challenge because it's there. If we can sequence more cells, it's going to cost more. It's still going to be a lot higher. I don't exactly know the number, but it's technically like the challenge that we can… We have the opportunity to sequence more cells that will cost more in terms of sequencing. It is a direction that like nevertheless is needed in terms of… Especially early target discovery programmes is a critical element. It is more manageable than the 10x Chromium per sample cost, the sequencing cost, it's still going to be high, like USD 100-200 per sample for sequencing, but it's still going to give you like a 60-70% lower cost compared to the 10x Chromium.

[[00:15:32]]
How would you say the commercial roll-out of high-throughput spatial imaging platforms alter or shrink the customers' budget allocation for traditional single-cell sequencing instruments, based on everything we've discussed so far?

[[00:15:51]]
I think the single-cell data and spatial, I mean they're complementary. I wouldn't say that like Xenium or newer version Atera is going to erode the need for single cell technology. There are a lot of challenges with adopting spatial technologies. I mean the cost is one element of that, but the throughput is another challenge, although 10x try to address that with Atera but the scale was just not comparable. The application is also different. For a spatial biology, it's a lot more used for the targeted questions and also like the diagnostics in the later-stage clinic. With some of the advances that in the biomarker space is happening with companies like Roche, entering into spatial companion diagnostics field, that's a different a set of questions that are being addressed by spatial technologies and single cell. I mean, single cell technology being cheaper will reduce the need for spatial or like a spatial going a little bit more high-throughput is going to affect the single cell. I think both will continue to grow, like the technology, and they have a different need basis, for the single-cell high-throughput perturbation screening, CRISPR screening or drug patient screening. I mean that's the biggest area of growth, but spatial is more on the, I would say, diagnostics and more on the late-stage clinic, which I wouldn't say they are very… They do overlap with that much.

[[00:18:25]]
How do you analyse the price elasticity of demand for single-cell consumables now that open source options are approaching the equivalent economics of the USD 100 genome milestone we've seen?

[[00:18:43]]
I think the plate-based technologies, again, it just lowers the friction. It's not going to resolve… All of these single cell technologies is processing the raw sample, it's not resolving the sequencing issue. I think that improves the plate-based technologies perspective because now the sequencing costs can also be lower. Again, it will support the idea that the large-scale data generation is feasible more than past. There is a huge need for data generation at the larger scale for training the foundation models and predictive models based on the perturbation experiments. I think it doesn't compete. I would say it does actually favour the plate-based technologies.

[[00:19:56]]
Why would a commercial biopharma end user continue to pay premium pricing for 10x Genomics' single-cell consumables when open source or lower-cost block alternatives are available?

[[00:20:10]]
I think there are a lot of reasons when we do design a study, like we still go with 10x, even if it's more expensive. There are a number of reasons because let's say if it's a clinical study that we are designing, the most precious step in that process is the patient sample that we get, and that they are limited. You do want to make most out of that. For those, we want to use the technology that is most reliable, most accepted in the community. Let's say, we have an external investigator, like academic collaborator that most often we do for those type of studies, I mean when they think of like a potential for publication, it's a lot more straightforward with 10x technology because of the community acceptance to use that. I mean we know it does have better sensitivity and specificity, but also it does make that process a lot easier. Although it's still a big chunk of the funding goes for that single cell in that study, we still would prefer that 10x over other technologies, 10x Chromium over other technologies just because that's the gold standard.

We do have limited access to patient samples and we want to get the most out of that. There is also historical reasons because most of the data that we have in the pharmaceutical or in the public domain in general, they come from 10x Chromium technology. There are integration issues, batch effect issues that we want to consider and a direct comparison, it's more easy to compare when all the data come from the same technology. Yes, I think there are reasons why the switch doesn't happen automatically. Over time, I think for a set of applications, it will be… Just because there are other pressures financially and the scale that will make basically the pharmaceuticals and academia to switch to plate-based technologies.

[[00:23:08]]
How durable would you say is the recent double-digit growth that has been seen in single-cell consumable reaction volumes if high-volume pharma accounts actively migrate their baseline screening to plate-based workflows? Would you say that increase in growth is a durable number that's likely to continue, or is that not a durable growth metric that you anticipate over the next few months or quarters or so?

[[00:23:51]]
Yes. I would say, it still will continue to grow. In general, regardless of the technology, if you say single cell or special, both are growing. There's more and more the need and the appreciation of the single cell and spatial and that the customer base is going to grow, regardless of the funding issues or changing from one technology to another, the customer base is growing. In terms of the instrument droplet-based or plate base, I think in the short term, we still continue to see that consumables for 10x technology is going to increase, not like it's going to grow, but it will eventually slow down. I wouldn't see it's going to be negative in the short term. It's a long-term problem from what I see. Still in academia or in the pharmaceutical, 10x Chromium is the gold standard. It's the to-go technology. Especially if it comes to clinical samples, I do see that it's going to take time to switch to other technologies. I mean other technologies also do need to improve their sensitivity and specificity for entering into those segments of the market per se.

I think in some way, there may be less overlap between what the 10x Chromium addressing vs what the plate-based technologies will succeed and the plate-based technologies will succeed in the larger-scale perturbation experiments, like high-throughput experiments generation. For their applications that… For 10x Chromium that will take time to switch to another technology. I would expect that this will continue to grow, maybe not at the same rate, but it's still positive.

[[00:26:43]]
What is your assessment of 10x Genomics' technological leadership and global commercial execution across the spatial landscape?

[[00:26:58]]
I think the problem is the preclinical spending in a pharmaceutical. When there is limitations in terms of the spending, the first thing that would get cut off is in the early stage like a preclinical discovery and more funding will go for late-stage development. We do see it to some extent in a broader pharmaceutical industry happening. It's a cycle. It will, again, return to early stage, there will be more spending in the near future in the preclinical development. Currently, we do see that challenge for favouring late-stage assets, and that's a challenge for 10x Chromium technology because most of the use that we have for single cell is either in academia or in early-stage drug discovery, not so much in late-stage development, mostly because, I mean, most… We do not have FDA approved endpoints for single cell is not the primary endpoint for many clinical studies that are being run. It's mostly like the good to have or secondary endpoints. I think that could create a pressure. Again, it's more challenging, if the funding in the early-stage drug development gets cut, that will affect the perspectives of the 10x Chromium.

[[00:29:22]]
How do commercial biotech and pharma customers currently balance their procurement between early target discovery tools and late-stage clinical assets?

[[00:29:35]]
I would say there is a shift toward more late-stage assets. That's mostly driven by some of the uncertainties in terms of like the patent expirations for many pharmaceutical companies that they currently face and also ensuring that they do have a sustainable pipeline. There is a lot more emphasis on ensuring that the late-stage assets gets commercialised and can sustain the revenue growth and the profit margin, which then it will flow back into like a preclinical or early-stage drug discovery. I would say, currently, we are in this stage that it favours late-stage assets over early-stage development.

[[00:30:47]]
How are upcoming pharma patent cliffs causing customers to extend their instrument procurement cycles, given that new platforms face strict CAPEX scrutiny?

[[00:30:59]]
I think that's a real challenge. Because of these challenges, it's very hard to make a case for, let's say, setting up a new lab or buying new equipments or new instruments because then we have to justify a lot of capital investments early on before any data is generated. For those reasons, there's a lot of challenges in making case for buying instruments, but on the other hand, we do have the potential for… There's more interest in utilising other CROs that specialise on running a specific technology. Let's say, if the new spatial technology from 10x is introduced, it's very hard for us to just make a case for acquiring the new instrument, but rather, it's more reasonable to go through CROs and using 10x technologies just for data generation externally rather than doing that investment in-house. That also, I wouldn't say entirely favours plate-based technologies for that specific reason, I mean, obviously, the general funding is limited and that does favour plate-based technologies but since there is the option to use CROs, like external data generations, where needed, we would still go for those technologies that require instrument even if they provide better data.

It's hard to compromise on data quality over financial reasons. Even if we are not able to make a case to buy the instrument, if needed, if the data quality dictates that, we would use external organisations to generate the data.

[[00:33:50]]
Why do you think there are slowdowns in commercial hardware purchases operating independently from traditional academic NIH [National Institutes of Health] funding headwinds? How did historical NIH funding headwinds trigger previous guidance withdrawals? How exposed is the core consumable baseline to another round of grant tightening?

[[00:34:28]]
I think there are multiple angles to this. In some ways, pharmaceutical industry is immune in the short term to those NIH cuts in funding. In the long term, because we do rely on… The internal data generation engines are not comparable what the public domain data generates, like the amount of data that is generated in the public domain. That is pretty much driven by public funding, by NIH funding and other sources of funding. The cut is going to show up itself in how much new data we have access to, and that may initiate more pharma spending because we do need certain data generated. If not available in the public, we may need to spend more in generating those data. At this point, in the short term, we do not see that effect so much. The effect mostly comes from the fact that academia is willing to collaborate with pharmaceutical companies more to generate the data, while in the past, they may have had more funding so a less interaction with pharma and academia would have been expected, but now there is more interest from academia to collaborate with pharmaceutical industry to generate new data.

In the long term, the life cycle of research, maybe in 2-3 years, we will see that, that's going to affect the… How much data is available for performing the drug discovery, and that will initiate more spending from pharmaceuticals to address the lack of data in the public domain.

[[00:37:02]]
What do you expect the average reduction in high-plex spatial fleet placements to be as pharma capital pivots away from validation tools and into active clinical pipelines?

[[00:37:16]]
That's very hard to put number in it. I mean, it's also a very dynamic process. It's hard to put actual number. We may see it through maybe… I mean some numbers that we see maybe… I mean it's going to be 10-20%. It's not completely like the shutting down, let's say, research in favour of like a late-stage assets. The reduction is real through not like a slowdown in hiring and also like in spendings. I will say, it's around maybe 10-20%. That's significant because there are a lot of routine assays that pharmaceutical industry uses. Often the first things that get cut are these novel technologies that cost a lot more and more exploratory.

[[00:38:40]]
For an installed Xenium customer, does Atera require a new instrument purchase, or is there an upgrade path?

[[00:38:54]]
I think from what I understand, it does require, but we are not… Again, because Atera is still… There is not a lot of data, it's not something that we would immediately switch, if it was made available to us, it's going to go through CRO, like I think that because there's not yet enough data for us to do a good benchmarking of the added benefit of the new technology. For us, the major challenge is not necessarily… If the upgrade is possible or not. I mean, certainly, if it's possible, it's going to make it a lot easier to adapt, but I think the major question that still need to be addressed is how much added benefit. We understand it's going to be more high throughput, it's like the whole transcriptome level analysis and it's going to make it more comparable with the CosMx technology. At this point, because there is not enough validation data out there, we are not in a stage of like that we need to adapt this as soon as possible. It's more like we need to see how community reacts and what data is being generated and if that data meaningfully adds value compared to what we get from Xenium currently.

[[00:40:54]]
If it's a new instrument, what's the typical refresh cycle for a spatial instrument in a biopharma setting compared to an academic core facility?

[[00:41:10]]
If the good use case exists, I think it's a very quick process to acquire the new… From my experience, less than six months to negotiate and get the system up and running. Again, it depends on really like if the need is identified, but with current cycle, I would say for many mid-cap pharmaceutical companies or small cap, it's going to likely go through CRO until a wider adaptation of the new technology is evaluated.

[[00:42:10]]
Is 10x Genomics' current instrument revenue collapse truly driven by Atera-related customer procurement pause, or is it an underlying indicator of macro capital containment across core labs generally?

[[00:42:25]]
Yes. I would say the second, I don't think it's specifically related to Atera. It's more like the spendings are restricted. There are many emphasis on… There's the challenges with fundings and starting the new study is very challenging. It does go through multiple rounds of discussing the fundings. I think it's mostly related to where we are in the economic cycle for academia and early-stage drug discovery in pharmaceuticals, not rather related to the technology itself.

[[00:43:31]]
What would you say are the realistic instrument ASPs and annual consumable pull-through commitments that are required across different experiment scales to justify an internal hardware purchase over a traditional outsourced service provider model?

[[00:43:46]]
I think for spatial technologies, for example, I mean, it's… Because they are very slow in terms of generating the data, we are limited by number of samples per month or per year that the machine can process. I think that's one of the biggest advantage or the new technology, the Atera from 10x is trying to address. Again, I think that will depend on… For example, with the new technology, they claim around 800 samples per year that they can process. With Xenium is a lot less, especially if we want to go, let's say, over 5,000 transcripts and do also proteomic. That's really like a lot lower. I mean maybe like around 100 or actually less than 100. That's for us, it's more like how much that instrument is being utilised. It's a lot easier, for example, for Chromium technology to make a case because that's the utilisation of that is very high but it depends on if it's… Even if we have the internal instrument, it's not going to give us more than 100 samples. Maybe CRO is a better choice because now we are not limited with the number of instruments that we have. I think it's the utilisation and I would say the scale that's going to make a difference in terms of what path is more reasonable.

[[00:46:06]]
Regarding spatial trade-offs, what core operational features differentiate a sequencing-heavy spatial assay from an imaging-centric platform during high-throughput tissue profiling runs?

[[00:46:23]]
I think the major differentiator is the scale and the throughput. I think the sequence-based technologies, like Visium or even like the GeoMx from NanoString or Bruker technology, their advantage is that you can do whole transcriptome. That's very good, especially for early-stage drug discovery when you are doing the target identification, for instance, or target validation steps or like MOA, like a mode of action of the compound that you are looking for like the off-target effects that we are looking for. I think the major challenge with the image-based technologies is like it takes a lot of time to run an individual sample. That's just based on how the technology works, especially if we go beyond the selected panel of genes or proteins and to go more unbiased with 1,000 or 5,000 genes, it takes multiple days to run that sample. That's the limiting factor. Although like we do get better resolution, it just makes it not a good option for unbiased early-stage target discovery or drug discovery in general, but it does make them a better tool for diagnostics that can have applications in later-stage clinic, especially if there's a smaller set of proteins or genes that are… Like targeted approach where we are not looking for the entire genome, but rather a subset of genes, but we do want better quality subcellular resolutions.

[[00:48:44]]
When deploying spatial platforms across multi-centre clinical trials, how do you think operational throughput constraints and data storage overhead limits lower the long-term ROI?

[[00:49:08]]
I think data storage is easier to resolve, but I think the challenge is the scale and throughput that is very difficult and it does create a batch effect and the variations between the different labs. I think that it create an issue with the computational pipeline to analyse the data. It is in a sense, the part of the technology that's going to be really the limitation of image-based technologies that we do need… Especially the sample handling will have an effect, but also the variations between different centres will introduce batch effect that computationally will make it challenging to reproduce or to assess, especially the clinical setting when we are limited with the number of biological replicates. Scale that's going to matter when we don't have a lot of biological replicates that centre-to-centre variability or batch effect will make the data less reliable. If we have a high throughput and instrument that can resolve the scale issue, will create more reliable data.

[[00:51:18]]
How rapidly do you expect the whole-transcriptome imaging scale of the Atera platform to accurately displace the sequencing-based Visium HD product line?

[[00:51:33]]
I think one element of it is the cost. I mean the cost might be cheaper. After the capital investment to do the image-based, at least to generate the data and the downstream might be more expensive because you do generate more data and it requires more specialised in data analytics. In terms of downstream cost is unpredictable, but… I think still the scale is a challenge. If that resolved the scale issues, I think there will be more interest to replace Visium technology with image-based technologies, especially that it does have similar ROI in terms of region of interest that is around that gives you the option like around centimetre area. I think we do need to see a little bit more data in terms of the reproducibility and the sensitivity and the specificity of the new technology and how it compares with Visium in terms of the data quality.

If those data are generated and the community starts to trust it, I don't know how much time that might take. My expectation is like it's going to still need at least 2-3 years because that's how that publication process, peer review processes work to get more data out. Then I can see that actually more and more people will switch from Visium to subcellular resolution technologies in image-based technologies. Even with Xenium, we do see more adaptation. We do see that actually more studies are preferring the subcellular resolution, especially that the cost is still high for both technologies, so we don't get the scale with Visium technology too. Both cost pretty much. Both of them are high cost in terms of… With Xenium, you at least get the subcellular level resolution.

[[00:54:23]]
Regarding AI computational biology and broader dataset monetisation, what would you say is the actual commercial addressable market and true monetisation timeline for selling high-plex spatial datasets directly to AI drug discovery companies?

[[00:54:42]]
The scale is addressed and the data generation for AI companies, it's going to be a huge market. I mean not only AI company, I mean every pharmaceutical is now leading in AI implementation. We are still in the early stage. The data itself is not valuable unless it is actually also utilised in training the predictive models for target identification, for MOA explanations, etc. If it does come together with the models, I think that also breaks through similar to what happened in the single cell that through perturbation screening that feeds into foundational models. I think the spatial data need also that breakthrough to… They do need the scale to resolve that, but they do need the breakthrough to interact foundational models to be utilised as a training set and for enabling the predictions. I think that requires resolving the scale issue, but I could say that becoming a big driver of the growth.

As you know, more and more AI companies that do not have the facilities or expertise to generate the data, but they are interested in utilising the data. I think that's the general theme around pharmaceutical and biotech. If there's a validated dataset that can be acquired, that's a lot easier process to make a case for and find fundings and to invest on rather than doing an end-to-end data generation to model training and doing the entire pipeline. I mean if the data generation is a reliable process and the good datasets are generated, I think that's going to be a huge market.

[[00:57:21]]
How do the long-term economics and margins of the AI data generation opportunity compare to 10x Genomics' legacy academic business? Do you think the company is better positioned to capture the market than its peers?

[[00:57:40]]
Yes. I think because they do have the customer base and with the acquisition of Scale, I think that now they are actually comparable with other competitors in the data generation realm. I think they are well-positioned because of their customer base. I think they are going to need to have validation datasets. I mean a similar route that the Parse with Tahoe-100M or other datasets like generated, I think it needs to be those validation datasets generated and start to propagate into foundation models and show that there is a benefit generated by training the models on top of those data. Then I think the trust that the community has on 10x will drive the market.

[[00:58:52]]
Why do you think reproducible data generation represents a more severe bottleneck for drug discovery AI models than the maturity of machine learning algorithms themselves?

[[00:59:07]]
Historically, in biology, it's very difficult to generate reproducible data. It's a lot of times, experiments are difficult to reproduce. They're conflicting each other, and that's the bottleneck that needs to be resolved. If the high throughput data generation allows more biological replicates, more technical replicates, more in-depth data generations, it will support that reproducibility issue. A lot of examples in pharmaceutical and in academic, in biology in general, that there is a very interesting observation, but that observation is very hard to replicate. If there is a larger scale data that the models are trained on, maybe we can trace it back to some exact mechanism that drives that observation and make more predictions that can be tested in the lab. Yes, like I mean, reproducibility is a general challenge in biology.

[[01:00:37]]
Thank you for that. This brings us to the top of the hour. I'd just like to close out by saying thank you for hopping on the line and all of your insight. This was a really great call. I appreciate you taking the time to opine on everything today. Hopefully, we can work together again in the future.

[[01:00:55]]
Great. Thank you very much. Thank you.

[[01:00:56]]
Awesome. Thanks. Have a good one.

[[01:00:58]]
Thank you. Bye-bye.

[[01:00:59]]
Bye.

Transcription ends at 01:01:01 of the recorded material