0:01
Thanks everyone for joining me today.
0:05
I will be talking about Aspect Analytics and how we are supporting researchers in spatial biology to advance our understanding of spatial biology, primarily on spatial multiomics data revolution.
0:22
I'll give a brief introduction of who we are, what we do, how we approach problems, and then I'll guide you through important aspects.
0:29
First of all, how we approach spatial multi-omics, what challenges we see what, but also opportunities, what problems you see when we start operating to support this from a scale perspective. From a very high level perspective, Aspect Analytics is the bioinformatics provider.
0:54
We don't have any wet lab components ourselves, but we do provide software platform called lead or data management analysis integration which can be extended by our collaboration computational teams.
1:09
And we provide hands on data analysis services, which tends to revolve around these new types of multi-omics integrations, but also increasingly all the data-driven QC.
1:21
You see, once you start scaling up, QC becomes increasingly important.
1:25
You make sure that every step of an analysis, certain assumptions are met.
1:31
In general, we have the goal of providing a unified platform solution that allows you to work with spatial biology.
1:39
Data comes from pretty much every assay, both in isolation and in integrations with others.
1:45
Right now we support the vast majority of commercially available assays needs, transcriptomics, proteomics or Mass Spec.
1:53
In terms of how we design the software, we focus a lot on enabling collaborations across the team.
2:02
In spatial biology, the teams tend to be large by nature, tend to be very interdisciplinary and what we focus on heavily is enabling non computational team members like biologists, pathologists to also be able to interact with the data directly while they're being supported by computational team members as well.
2:21
Another tenet that we focus on is FAIR and reusable data because we're generating a lot of data, but we also have to make sure that we embed into the maximum potential beyond the scope of the initial project in which it was generated.
2:34
And finally, which I'll talk about today mostly is spatial multi omics and bringing that into high throughput.
2:41
I've listed some of our collaborators below and you can see that it ranges from big pharma users in the fields all the way up to meeting academic institute.
2:49
So we cover the whole spectrum essentially in terms of where we sit application wise in the pharma pipeline.
2:59
We start very early on in target ID, target validation.
3:04
But we're also very active in preclinical work for PKPD type studies and increasingly in clinical trials for patient stratification where spatial biology will be used to identify patients of interest or molecular or spatial signatures.
3:21
In terms of therapeutic areas, we cover all the main ones that you see in the fields ranging from oncology, immunology to a lot of neurology work.
3:33
They're also more use cases here that we're not showing, but these are of course the main ones.
3:39
And in terms of assays, pretty much everything that is available in the field.
3:43
So spatial transcriptomics, whether it's imaging or sequencing based, minibulk or subcellular spatial proteomics, whether it's based on fluorescence microscopy or mass spec IMC maybe. We are increasingly working on bringing in LCMS as well where you start from a laser capture region.
4:03
I want to overlay that information.
4:05
And we also support an assay that is somewhat more exotic in this conference at least, which is Mass Spec imaging, which allows you the image metabolites, lipids, peptides, glycans in tissue.
4:17
I'll show you use cases that as well.
4:23
Our products, the weave platform is a cloud platform where essentially the various spatial biology data sets you may have, including traditional histology, but also other modalities like single cell RNA seq or bulk proteomics can be uploaded into one managed platform, which is formally certified for information security, business continuity, etcetera, where you can then analyse, integrates and work with your data.
4:52
In general, we do this in support of multiple teams in pharma, where the main teams are of course, the project teams generating data to ultimately answer some scientific hypothesis.
5:04
But we're increasingly seeing collaborations within pharma with AI teams that have a different take on this data, different priorities, different requirements as well in terms of how we can use it.
5:17
And lastly, because data generation is picking up, the volumes are also picking up, the investments are also picking up.
5:24
We want to make sure that data's reusable.
5:26
So we see increasing involvement from internal data governance teams that focus heavily on making data FAIR, accessible, reusable.
5:35
So all of the features that we work on are also related to this.
5:38
The platform itself has an interactive web interface where we can show demos at our booth in the exhibition hall, but it also comes with the programming API to enable computational and reuse to extend it and even bring in their own pipelines.
5:55
In terms of features, I won't go over all of them, but we focus a lot on data visualisation, data management, making data reusable and really bridging across everyone in the team that is involved in analysis.
6:09
When we talk about spatial multiomics, we start typically from what we call a stack of serial sections, each of which tends to be analysed and using different assays.
6:21
Sometimes you're able to combine multiple readouts from a single section, which is good.
6:26
It makes life easier.
6:28
But for a lot of assays, that's just not possible.
6:31
So we end up with different tissue sections that were analysed, different data formats, different spatial resolutions, different content of the data, of course, but ultimately we want to consolidate that into one measurement that you can analyse as one object.
6:49
That's the goal, but it comes with various challenges.
6:55
Within our platform, we've developed a process called stack fusion where we take this set of measurements and we consolidate that into one integrated data structure for computational people think and data object or a data frame where you now have the same observations and then all the different readouts combined.
7:14
So you can analyse it now as if it was one measurement.
7:18
Of course, you can then do joint visualisation, but you can really do any kind of downstream analysis you might be interested in, whether exploratory or targeted.
7:27
This is what we like to call true spatial multiomics because now we're not limited to whatever one single assay can measure.
7:35
You can really combine information from any set of assays you might be interested in to answer your scientific question.
7:43
To make this concrete.
7:45
We have a few examples.
7:46
The first one is work that we did together with the team of MD Anderson, Jerry Burks and Sammy, who also presented here yesterday, and Erin Sealy at Texas, which was about high grade serous ovarian cancer, which is a very lethal malignancy of course, and it accounts for the majority of ovarian cancer deaths.
8:07
The problem with ovarian cancer is that a lot of patients relapse after the primary intervention which consists of surgery and chemotherapy.
8:17
The goal of this project was to better understand minimal residual disease for MRD and particularly to look for biomarkers that may be of MRD or how chemo resistance comes to be.
8:33
In this project, this was a small pilot that is now being scaled up.
8:36
We had access to the 8 samples, 4 MRD negative and the stack is shown here in the middle.
8:42
So we have two sections.
8:44
The first one analysed using the old school visium assay for spatial transcriptomics with an accompanying H&E.
8:52
The second section using spatial proteomics, in this case the common instruments of Lunaphore.
8:58
And then this third section using mass spec to target in this case metabolites, glycans and peptides.
9:05
So if you look at this from an omics perspective, we have 5 different layers, transcriptomics, proteomics, metabolomics, glycans and peptides some H&E in the mix for good measure.
9:18
And we want to bring all of this together, right.
9:21
So what this involves and the first step is image registrations.
9:25
We have to establish a common coordinate system across all these different measurements to be able to translate ROIs and compare data points etcetera.
9:36
And this is done through registration, which our platform provides many capabilities for ranging from automated ones to manual ones depending on which types of combinations you're trying to register.
9:48
What's important in registration is that everything is long reached.
9:51
So even if you work on the same section, you may have complex tissue deformations that we need to account for.
9:57
So it's not just a rotating and scaling the sample, It's more complex than that and that's really important.
10:04
We have multiple registrations to take care of.
10:07
I won't go into detail on all of them, but we always in this project, we combined everything to the level of the H&E.
10:13
For the mass specs of the bottom most section after registration of the comments, we can overlay that data.
10:19
Similarly for the mass spec data, we can also overlay that after registration.
10:24
What's interesting here is that you can see differences in FOBs, which are very common across different assays, but also differences in spatial resolution.
10:33
The mass spec data here has a pixel size between microns, which is much lower spatial resolution than microscopy based assays of course.
10:42
So after registration, we can now jointly visualise data, we can interact with it, but we still don't have what we call a structure for which we still have to decide what are our roles in the integrated data structure going to be.
10:56
And for this we used the individual pixels of the mass spec data.
11:01
This is a choice that you make in a project, so you're very flexible. Here you can see that these pixels are larger than individual cells.
11:10
And through this aggregation we now have for each Row the cell pipe proportion from the proteomics data.
11:17
We have fluorescence markers, we have peptide, glycan, metabolite information and we also have transcriptomics coming from the Visium assay all in one data structure and this allows us to do various types of analyses.
11:29
So we can now find correlations across these different types of omics classes.
11:36
We can stratify this like within MRD positive, within MRD negatives, even within pathologist annotations you may have on the samples.
11:45
We can do that for all these different pairs of course.
11:47
And so here we're showing metabolites versus glycans, metabolites versus glycans that you can really do anyone wants.
11:54
The second type of analysis that we like to do is in this case defining glycan peptide or metabolite signals that are specific to certain regions in the sample or even specific cells.
12:08
Like in this case, we have two more cells in green, calves in yellow and in new cells in blue, and we can then find molecular signals that are not accessible via antibodies for those cell types by combining the different results.
12:24
The third type of analysis we give in this project is a very spatially focused workflow where we looked at the tumour stroma interface, where this interface was essentially localised via polyomics markers and pathologist inputs from the MD Anderson team.
12:39
And essentially what we're looking for is what changes as we move in this boundary.
12:43
Changes could happen in terms of cell type density, molecular expression, multicellular environments.
12:51
Even in this table here we're showing changes in terms of cell type density.
12:55
And we can see for example of CAFs are mainly localised at a very specific distance to the tumour, etcetera, which may have of course all kinds of implications if we look at a different project that we did together with A star in Singapore.
13:10
This is a team of JPS Young.
13:12
We integrated data acquired from the same section.
13:15
So they were able to do Xenium first with 290 gene panel, then COMETs with 40 markers and then H&E all on the same section.
13:27
In this case, what we can do is integrate everything up to the level of the individual cells.
13:33
So now we end up with first cell, both transcriptomics and proteomics readouts, and I'll show you how we did this.
13:41
The process is the same as before.
13:43
So you have to start with registrations and then you have to fuse everything to the individual data points.
13:50
If you work on the same section, registration has to be spot on.
13:54
And what we're showing here is the H&E on the left.
13:58
And then would that be signal from the Xenium assay here on the right?
14:01
And on the bottom, you can see the other bit and you can see that there's a 100% match nuclei, nucleus to nucleus across these modalities.
14:10
And this is true across the whole sample.
14:12
So if you're interested, we have a live demo of this data set at our booth.
14:16
And this is really key.
14:18
And again here, it's not as simple as just rotating and translating one of the images because we do have complex tissue information, especially with cyclic IS assays like the COMETs, the tissue will swell in various ways.
14:31
We need to be able to account for that.
14:34
Once we've integrated things, we can now look at the data in a multi-omic sense.
14:38
So we can see the proteomics data and the transcriptomics data and the H&E all together.
14:44
We can look at cell segmentations coming from different assays for together.
14:49
So this is if we work on the same section 1/3 type of application that we see a lot in our work is in preclinical work is in situ PKPD studies where we use mass spec imaging to localise the drug and drug metabolites in tissue, both the active compound with also other metabolites that may be relevant to see first of all, whether it reaches the target cells for instance.
15:17
Whether my drug crossed the blood brain barrier as a simple example.
15:21
But also are toxic metabolites being formed?
15:24
If yes, where are they, where do they go?
15:26
Do they stay? These kind of questions.
15:30
But since we now have spatial biology available as well or new assays, we can not only localise where the drug is using Mass Spec, but we can also look at what the phenotypical effect is of the drug in situ.
15:44
And we do this by bringing in spatial proteomics, spatial transcriptomics to see what the effect is of the drug in tissue.
15:52
We can look at what the immune responses for instance after a vaccine injection at different time points, we can start assessing mechanism of action of the drug etcetera.
16:03
That's the first use case.
16:04
Second thing I want to speak about is moving to high throughput.
16:08
So when you work in high throughput, you may have a process that works at small scale, but that starts to break down when number of samples increases.
16:18
And this is what we aim to avoid, right?
16:20
So when we move into high throughput, ultimately we don't want to compromise the depth of analysis simply because we don't have the time, right?
16:30
So this brings a number of considerations that we focus on heavily in how we approach both projects, but also the design of our software platform.
16:39
The main one I cannot overstate is the importance of streamlining collaboration across the team.
16:45
So that means having a centralised place where everyone can do their work, can communicate efficiently and we create tight feedback loops.
16:53
So what we see in many projects is that it sometimes takes a long time to get feedback between computational and non-computational team members because transferring data is difficult, it's hard to access and so on.
17:09
Within Weave, you send a link, someone else adds the comments directly there, most software necessary.
17:15
And this really enhanced feedback loop.
17:18
It also means generally what we see is that team members not only communicate faster, they also communicate more, leading to better results, of course, on your data volume increases.
17:29
You also have to be able to actually run the analysis, which has a computational component to it.
17:34
We need to be able to run, let's say 50 cell segmentations in parallel.
17:39
But it also has a component of standardisation.
17:42
So when you work at scale, you have to have the set process that you follow for every sample where you have a lot of QC points along the way.
17:52
Because when you operate at scale, failure will happen, samples will go wrong for various reasons.
17:58
You need to be able to detect and deal with it efficiently.
18:03
In terms of data analysis, what that also means is that we track provenance.
18:08
So how was a certain computational result established?
18:11
What were the inter parameters?
18:12
Is this the correct version of the result that someone else can work on later on?
18:16
These things are really important when you start building a lot of different data.
18:21
And finally, what we tend to see, and this is an increasing focus that we are happy to witness is the importance of data news.
18:28
But we don't just want to analyse data in the scope of the project in which it's generated, but we want to create essentially an internal knowledge asset across projects.
18:38
And this means, first of all, the data has to be very accessible.
18:42
And we approach that we'll leave by making everything available in open formats.
18:54
But also the data is FAIR.
18:56
This is about data management.
18:58
So you need to be able to specify sample information in the platform where you store the data and you need to be able to link with ontologies.
19:05
So you have shared vocabulary if you want to link different studies together, etcetera.
19:10
All of these things are available in these.
19:15
One type of project where we see data reuse and data consolidation as very important is international consortium projects that we are very often a data backbone of.
19:28
I'm using Gliomatch as an example, which is a horizon projects focused on glioblastoma and high-grade gliomas.
19:35
In this project we have about 10 hospitals generating clinical data, MRI microscopy.
19:43
We have two core facilities doing station biology and all of them have an automated way to upload their data into this into the lead platform that is used in this project where then other stakeholders in the Consortium can access the data by physicians, biologists, etcetera.
20:01
So we take care of all of the logistics.
20:02
All the logistics in cultures like these which tend to be non-trivial and here it also helps the fact that we are compliant and formally certified to make sure that this is done using best practises.
20:18
So finally, in conclusion, I've introduced Weave, which is one of our products for spatial multiomics, which you should see as a foundation for data management, analysis and visualisation, which has end to end capabilities to integrate data from different assays acquired in serial sections.
20:39
And right now we support most commercially available assays.
20:42
If you happen to have one that we don't yet support, we'd be happy to work with that data as well.
20:48
Next, in the platform, we also provide hands on services, which tend to be the more high-end bioinformatics work where the existing teams may not have experience with certain assays or certain combinations, but also QC as an action. QC tends to be quite process specific.
21:09
So how we work in the web lab may dictate how you see certain results and those are things that we developed together with your team to ultimately be able to deploy it in a flexible way that works for your team.
21:24
We also integrate very deeply with vendors in the space, make sure that's onboarding data they generate into our platform is quite easy.
21:34
So in conclusion, we focus on integrating data across different platforms, doing that in a high frequent manner where you have limited time available per sample, but still won't run the same depth of analysis.
21:48
And we focus a lot on communicating across stakeholders, both within project teams, but also in terms of reporting to downstream stakeholders that we can see.
21:59
So that's it from my side.