Episode Transcript
[00:00:03] Monica: Welcome to UVA Data Points. I'm your host, Monica Manney. In today's episode, we're diving into a fascinating intersection of cutting-edge science and data innovation. As technology continues to evolve, researchers are increasingly turning to brain organoids, miniature lab grown models of the human brain to unravel some of the most complex mysteries of neuroscience.
We're joined by three brain organoid experts. Thomas Hartung, professor of environmental health and engineering at Johns Hopkins University. Jack Van Horn, professor of psychology and data science at the University of Virginia. And Lulu Jiang, assistant professor of neuroscience, also at the University of Virginia. Together, they'll shine light on how brain organoid technology is reshaping our understanding of the brain and how data science is playing a crucial role in unlocking its secrets.
[00:00:51] Jack: So let's get started. Tom, I would love to hear about the basics of brain organoids, what they are, why they're generating so much excitement right now, and what makes them a compelling model for studying human brain development compared to say, traditional systems like 2D cultures or even animal models.
[00:01:08] Tom: Sure, thanks for having me. You know, cell culture is a difficult beast, especially if you want to do human cell culture.
I work on the brain, so if I want to pick your brain, and I mean it, you will not be very forthcoming, I assume.
And this is the big problem. Good human cells of the brain are difficult to get.
So we have been working with rat and mouse brain cells for many decades.
And what turned out is that to really get a functional type of culture, somewhere interactions of different cells and active networks can be studied. You better go 3D, you better have the glial cells available.
Actually, it was already in 1979 that Honaker in Switzerland, in Lausanne developed a method of brain organoids from red cells. And this was a Nature paper. So it shows you that it is, was found to be meaningful already then. And we did adopt this model in 2002 because we wanted a good model of the, of, of the brain which was more than just keeping some neurons alive and measuring some cytotoxicity or so.
When I moved from Italy, where I was doing this work, to Hopkins in 2009 to assume a chair for evidence based toxicology here I brought two of my former PhD students as postdocs and they established this.
One of the first thing we started was to humanize this model because in the meantime the technology of induced peripherent stem cells gave us the opportunity to create human brain cells. And we adopted the protocol of Honager and brought it together with the technologies of Yamanaka and others. Who were developing induced pluripotent stem cells.
It was so tight. We were not the first. Lancaster and others published in August of 2013. Our first publications from December 2013.
Being a pharmacologist and toxicologist, my interest was in a highly standardized model. One where not every brain organic looks different because I need something standardized to test on.
We were the first in 2016 then to describe the mass production using bioreactors in order to produce large amounts of these brain organoids.
To give you a feeling, I assume most of your listeners are scientists somehow and have heard about cell culture. A six well plate gives us about 4,000 highly standardized brain organoids. This opened up for us to study a number of diseases, a number of things, and also biological computing.
[00:04:05] Jack: Thomas, have you seen any breakthroughs from your own work where organoids have captured something that just couldn't be modeled before?
[00:04:11] Tom: Absolutely.
A very prominent example was that we were using them to show already in May of 2020 that SARS CoV2, the virus behind COVID19 is infecting human brain cells. The first to do so. And this was really a big hit. Unfortunately, we had to report human brain cells are being infected. And this was because we had a model, a human model available and colleagues next door had the virus. And it was very easy for us to show this.
And this shows you already that having a human model is really critical. And this becomes also critical in pharmacology at the moment, because 57% of the new drugs are actually either human proteins or antibodies against human structures, which means you cannot test them in animals. You need something human.
Roughly a quarter of all clinical trials are done on brain diseases.
Imagine how important it is to have a human model available. And even if you're not studying the brain or brain disease and trying to develop this to de risk substances to show that they are not making people dizzy or are impacting on brain cells is a critical thing of drug development. So these human based models have enormous prospects here.
[00:05:35] Jack: I like this idea of again the human having a human analog that could then go into say animal models, which could then go back into the human. It adds that layer that a lot of clinical trial or a lot of things kind of need just in order to perfect them along the way. So Lulu, I'm wondering how close we are to creating organoids that are true mirrors of the complexity of the human brain, Whether in terms of the numbers of including multiple different cell types. I mean, Thomas mentioned glia, any like, as they develop, do they develop like cortical layers, for example, or they develop like functional connectivity which could then itself be modeled in terms of not only looking at say drug responses, but also the functional reaction to those drug responses. Can you tell us a little bit about your work in that area?
[00:06:28] Lulu: Yeah, that's very cool. Like Tom mentioned, the beauty of these 3D dimensional organoid structures to recapitulate the neuron interact neuronal interactions, the circuitry, the neuroglia interactions and also form the synapse and also all those well refined structures like the human brain. In our lab we study Alzheimer's disease and dementia related disorders which is very important to make sure this organoid are mature functional, can have all those structures between neurons, glial cells, especially glial cells. Because we know inflammation in the brain neuroinflammation is a very important driving force for the neural generation in Alzheimer's disease process.
We really tried harder to incorporate the astrocyte microglia oligodenary site, all those glial cells in organoid system so they can form interactions with the neurons. So we can study the cell cell chat. Also the drug response or disease progression with each individual cell type. Contributions in our lab, Tom mentioned about Lancaster published the first cerebral organoid back to 2013. Those organoids show the different cortex structure. But in our study we successfully incorporate microglia by differentiated the same IPSC stem cells into neurons, progenitor cells, astrocyte progenitor cells and hematopoietic cells. If we look at the brain development actually microglia, which is the immune cell of the brain, have the different origin. Those cells develop from the yolk sac instead of the neuronal tube. That's why we came this idea. We will differentiate them into their right track first. Then incorporate them together into organoid system. Now our lab have advanced the model with the right ratio of neurons-to-astrocyte microglia at 45%. 45%, 10%.
The second generation of organoids we generated, we also successfully incorporated oligodendrocyte which forms myelination, the neuronal sheath to protect the neurons and form the synapse between neuron glial cells.
Advancements we have been using and with these advanced models, we successfully generated the Alzheimer's disease model with the amyloid plaques present in organoid and also the neurofibrillary tangles which is the hallmarks of disease progression in Alzheimer's disease.
[00:09:06] Jack: What's on your wish list for organoids? What are they still missing that you'd love to see incorporated into their development? That will help augment some of the work that you're doing?
[00:09:17] Lulu: Yeah, that's really good question.
Actually working on generating the vasculature for the organoid because we know blood brain barrier is very important and blood flow is also very important for the brain function. Also in Alzheimer's disease there's a particular subtypes of dementia called vasculature dementia. Those patients have the circulatory deficits first and then contributed to the neurodegeneration causing dementia.
Now we are working with Dr. Don Griffin’s lab from the UVA School of Engineering. So we use scaffold to form a vasculature surrounding the organoid. So the future direction of one of my wish list is to add epithelial cells, endothelial cells and vasculature to to this organoid system.
[00:10:08] Jack: Wow, that sounds so exciting.
Tom, you were talking about the role of organoids in pharmacology and immunology. I'm just curious how brain organoids are really changing the landscape for drug discovery and for pre clinical testing. Do you think they're living up to their promise yet or is there more to go?
[00:10:28] Tom: I mean the limit is the sky, but we have already come pretty far I have to say.
And what holds here for the brain holds in more general terms.
My wife, Lena Smirnova and me, we have been pushing the entire field by creating an international society for microphysiological systems and creating a series of world summits. We organized the first four of them ourselves and just finished in June in Brussels. But with respect to the brain, it is really quite remarkable. And when we started, we expected we have to produce different subtypes of brain cells separately. So we were looking into protocols for astrocytes and neurons. And then we were blown away that our protocol, they simply produced all of these glial cells, even the oligodendrocytes, which are very rare in human cell cultures. We found that 40% of our axons are myelinated, which gave us a lot of opportunity now for demyelination remyelination experiments.
And this was already in the 2016 publication and very similar, we're on the same page what Lulu is doing. We have been incorporating microglia.
The difficulty here is to find conditions where they are stably integrating. In many of these models you can add them, but they disappear after a few weeks.
We have now conditions where we can stably integrate them because there's no brain disease without an inflammatory component.
And it is really critical to have good microglia in these models.
And very similar. We are interested in patient cells. So we have been looking into a number of different patient populations, mainly autism, because this was one of the driving forces for us new developmental diseases.
Because I also always felt this is more or less a fetal model of the brain. And it seemed a bit odd in the beginning to look for neurodegenerative diseases in something which is not yet born by the state of the brain. So we have been looking some rare diseases and also in autistic lines, but lately we have done just the same. We have been producing them from amitraphic lateral sclerosis, from Alzheimer disease and others because they really show some of the hallmarks of these diseases.
Enable us to study things which you cannot use study in animals.
[00:13:03] Jack: I think this is just so exciting and I like the, the, the whole thing where you're seeing, Thomas, you're seeing them just the glial cells for example, just happening as a consequence. But Lulu is able to engineer them. And I like the, this idea of being able to like, you know, have this like dial you could turn to basically include more or less kind of cell to cell ratios is really kind of exciting. And I'm particularly interested in autism, Thomas. So that's, we have, we have a shared interest there.
At the same time, I'm also interested in myelination and myelination in autism. And so I could see these organoids being an interesting model system for a developmental condition like autism where it looks like there's some alterations in myelin that you could explore using the organoids. So, so I'm curious, Lulu, also, if you could maybe also tell us a little bit about your modeling of Alzheimer's, for example, and maybe some of the more exciting or surprising ways that organoids have been used to model them.
[00:14:08] Lulu: Yeah, the way we use organoid system for long term goal, we try to develop into personalized or precision medicine because Alzheimer's disease, there's a lot of phenotypes, different patients, different populations have different symptoms and manifestations.
It's a very heterogeneity disease. We try to understand why some patients sensitive to this particular treatment, but some patients actually not sensitive at all. We now use this organoid system to study the genetic risk factors for AD. For example, APOE4, which is the predominant risk gene for AD in the traditional way, like mouse models, 2D cultures. It's difficult because APOE4, this protein itself has a very different structure structure from human APOE to rodent or mouse rat apoe. All those animal models actually were difficult to translate into future studies or clinical trials. That's why the organoid is very important now we use a patient derived stem cell, an iPSC cell, with APOE4 risk gene. Now we engineer that into the organoid. Now we use that organoid system to compare to the APOE3 which is the common variant of APOE gene to compare. See why this risk gene is driving disease progression. What is a risk gene? That's something cannot be done by animal models. I think this is really advancement and also innovation of organoid application in the disease studies is surprising but also really important and significant for long term.
[00:15:49] Jack: Here's a question for both of you. Maybe Lulu, you could maybe start with. This is about the variability that is batch to batch or lab to lab variability.
Whether or not there are any standards in the field that people are following. As organoids become more ubiquitous, what are people doing to make brain organoids more standardized and hopefully reproducible?
[00:16:12] Lulu: Yeah, that's a good question. I think Tom also mentioned that we really need a model standardized so we can, we can repeat it from batch to batch and from lab-to-lab. So the the first generation of the traditional cerebrogno, the Lancaster group published back to 2013. I think those organoids sometimes larger, sometimes smaller. So there's variabilities in our lab we try to generate this organoid with standardized size between the dimension of this organoid is between 300 to 500 microns.
The size of those organoids is very standardized. So there is no death curve. Because when the organoid is very large, sometimes there's death core with lack of nutritious factors, neurotrophic factors in the center. So our organoid we use Agriwell plate which is new method so we can mix the cell type first and put them into the agriwell. Those Agriwell with 300 small wells on the bottom. So those 300 organoids will be sisters and brothers. They are exactly the same with same same cell number, same ratio of different cell types. So that's the way we standardize it. I think that's good for drug screening and study of different genotypes and comparison from batch to batch is much more standardized and stabilized compared to the previous model of cerebral organoid.
[00:17:43] Jack: Thomas, what do you have to add there?
[00:17:46] Tom: I mean first of all, I agree completely that this is a big advance from the respective animal models. And this is close to directing centers for alternatives to animal testing on both sides of the Atlantic.
And it is really opening up for human relevant research.
As I said, our work came out of the autism research. We were interested why these numbers are increasing. And while a large majority of the field believes that's genetics.
Genetics do not change that fast. Yeah. We cannot explain these dramatic increases alone by genetics. And what we have been doing is looking for the susceptible individual which is hit at the right time point by a chemical. So the, the genes load the gun but the exposure pulls the trigger. Yeah, that's, that's a, that's, that's an interesting rule. And we were very excited that in 2020 we could show the first pair of a risk gene and a risk chemical acting synergistically and producing relevant phenotypes. In this case, SHOX, which is a risk gene responsible for about 2% of autism cases.
And Cloperifos, a pesticide, which was around the time banned by the EPA because of its likely liability for some cases of developmental neurotoxicity. So you really can do, as Lulu said very nicely, you can do things you cannot do with animals. Yeah, and that's, that, that's really a very important step here.
And we have, we are working with very similar sized brain organoids. From our experience we don't go above 500 micrometer. So that's the point of a, of a sharp pencil or a mid size grain of sand.
So you can see it, you can handle it. But this is sufficient for most experiments. So when I'm saying I produce 4,000 of these in one six, well plate, you can imagine how much experiments you can run by this. We are avoiding the problem of the original protocols by Lancaster and so which were large, beautiful brain organoids, but rotting in the inside.
As a pharmacologist and toxicologist I cannot use these.
And we're using duratory shaking to produce perfectly round brain organoids.
But this also means that we don't have really any sonation in our model.
It's not at all specialized. It has all of the targets, all the type of neurons we can produce. Physiological level of glial cells, but it does not have functional.
Different functional units is more like a flatworm organizationally and by the number of neurons we have.
[00:20:46] Jack: What's the thinking of the international community of brain organoid researchers? What's their take on methods to standardized or basically reporting maybe to help ensure that studies are reproducible.
[00:21:03] Tom: That's a big issue. I would say that the community has not really formed. There is too few commercial providers of these systems which would, let's say, seek industry standards, which I think are really necessary.
There is a few opinion leaders in this field which have been dominating the research, but they come with quite different approaches.
I've been heading for the European Commission, the validation body for what they called alternative methods to animal experiments. So I was very much into quality assurance and for this reason we have been developing first good cell culture practices, how to do experiments, and we extended these in 2022 to microphysiological systems like the brain.
We have also been publishing on quality standards for different types of organoids, including the brain. And actually right now while we are talking, I just saw an email popping up that our proposal for reporting standards has been published.
Which is saying this is a proposal now to the community to discuss for the stakeholders, this is a checklist of what we should report if we are publishing on a microphysiological system, on a complex in vitro system.
Because I think a lot of the irreproducibility in our field comes really from not simply not being complete in what we are telling others about how to do the research.
[00:22:42] Jack: Absolutely, that comes up again and again. So let me now bring in data science and kind of your computational analytics into this conversation. So brain organoids experiments, as we've just discussed here, they generate a lot of organoids, it generates a lot of data and I presume that you're able to analyze single cell RNA SEQ and electrophysiology, maybe even doing some three dimensional imaging. How are we managing and making sense of this all? And maybe I'll start with Thomas on this. And what analytic approaches have emerged as the go to statistical approaches?
[00:23:17] Tom: Yeah, that's, that's again a big one. You know, I'm, I'm very much known for my cell culture work and which ultimately led also to the brain organoids and all of this quality assurance and this past stuff. But I have handed over most of this work on brain organets to my wife Lena sma, who is professor in our team. And we are working together very closely.
And I have been focusing for the last 13, 14 years on artificial intelligence because then two people joined me, Tom Lichtenfeld and Alex Mertens, as PhD students. But they came with machine learning backgrounds and I dared to supervise them because I had been in the medieval times before and the beginning of my studies I've been working as a coder and have been studying also mathematics informatics for three and a half years.
And so I dared to do this and this positioned us very early in artificial intelligence and data analysis type of things. Big data require AI. Yeah. And, and some seven years ago I became the founding chief field editor of Frontiers in AI where we have published now 1,400 articles. So I'm really looking a lot into this field and the most interesting development out of this was that during the pandemic Lena and myself, we were sitting at home as everybody else and we were saying okay, what can we do now with this interest in brain and in artificial intelligence? Can't we do something together again?
And this was the starting point for what we call organ intelligence. OI maybe are using electrophysiology microelectrode areas to connect essentially a computer with the brain organoid and start talking to it with the idea that we can start producing cognitive simplest forms of cognitive functions, learning and memory.
And we are following approaches from our. Also our collaborators in cortical labs. Let these organoids play little video games.
[00:25:29] Jack: Yeah!
[00:25:30] Tom: Pong or little jumpers and yeah, so this is. It is really science fiction. Yeah. But it is working. Yeah. And so we see that these systems probably have been bought to death all their life. Yeah. They wanted input and they didn't get something. And now we're giving them input and feedback about their action and this. They enjoy it. Yeah. I have the impression, I say it's impression because I've not proven it in peer reviewed papers yet. That really alone the fact that we are engaging them in some tasks is leading to a more differentiated and functional organometry.
And this is data rich. Yeah. I can tell you we're producing gigabyte of data in, in one experimental day by scaling it at the moment. And we work very closely with the, with the applied physics laboratory. We have a enormous intelligence unit, 120 people and using fantastic tools, state of the art in analyzing these things. So we have a very interesting area of work and we now hope to create not only the community, a website, a journal, a bulletin we're sending out on augmented intelligence. But we hope to do next year the, the first conference on Augment intelligence.
[00:26:50] Jack: Lulu, what sort of analytical frameworks are you exploring and you know are. Is organoid intelligence on your list?
[00:26:58] Lulu: Yeah, the data science is very important for organoid work. So we, we mentioned that these are analogs of the human brain. But we wanted to make sure these organoids really recaptulate the human brain. So we cross reference to the human brain data, the organoid we are developing to study Alzheimer's disease. We try to develop amyloid plugs, neurofibrillar tongles which is the tau protein pathology.
But that is real to human brain or not. So we compare it with the human data set. We have Dr. Aiying Zhang from the School Data Science. AI is the expert in analyzing imaging data. So we have all the PET imaging data from the human population with the live patients cohort. So those cohort of patients we have the brain imaging data from the early stage until later stage. We are still following up as a longitudinal study in that cohort of patient will compare their pathological development of amyloid plaques, tau pathology and compare those to the organoid we are generating. So far our organoid really recaptulated the timeline and the progression of of the disease hallmarks.
This is the first point of data science really important to gather the human population data set compared to the organoid. The second point is you mentioned the single cell RNA sequencing. That's important too. We wanted to know which cell type contributed to disease progression at each individual stage of the disease from early stage until late stage. We used a live organoid to study their single cell arm sequencing at each time point.
With those data collected, we can analyze the progression at each stage. What is the early stage event? For example, probably the neuroinflammation, probably the mitochondrial damage or probably other genetic factors or transcriptional changes.
Far we use organoid system found that IGFBP pathways, those population of proteins is the early onset event in the disease progression which can contribute to the APOE4 population in their disease progression. Also we use the electrophysiological analysis to monitor the function of those neural activity, the firing of neurons, the neural glial interactions and synaptic activities.
Those electrophysiological data set are also pretty big. We involve the data scientists to help us to analyze those data.
After we get the single cell neuron sequencing data, we cross reference to those transcriptome data generated from postmodern brain tissues.
We know recent years we have nice publications on nature cell science about the postmodern brain tissues from 80 patients. Though those data set are very valuable. But those patients already at very latest stage, our data set can compare to those data set. So we will know what is the milestone in this disease progression.
But in addition to that, our organoid can also predict from the early stage because we can collect the tissue at very early stage and compared to the late stage.
Data science is really important in all these data set, all these analyses we wanted to compare side by side from the organoid to the human patient result.
[00:30:39] Jack: This has really just got me thinking. This is just such a provocative thing. Especially as what Thomas was just talking about is really the combining of these different data types. Whether it's spatial multiomics or you're using imaging, or you've got physiological signals which are varying over time or things that you can basically pinpoint when an early insult might happen and then watch the consequence for later. This is just so incredibly powerful, but it's also like really complex. How do you combine all of that data? Even when you've got these organoids are right in front of you, how do you combine all that data? And I'm curious here. If one were able to do this successfully, is there a potential for providing, for lack of a better term, a computational rendering of what might be like a digital twin of a organoid, which of course would then be a sort of a digital twin of the human that provided the cells that created that organoid. Basically an in silico computer version of the thing we're interested in, basically based on all that data collected together to produce an almost organoid intelligence based upon the physiological data that you're obtaining. Do you see that? That is a potential thing and where data science may be able to fit in with that.
[00:31:56] Tom: Just wanted to say these organoids are highly complex things, and the more complex you make them, the throughput goes dramatically down.
It is going down to an extent that sometimes the experiments on animals are sometimes cheaper and less work to do. So this is really important that we find ways also by at least in part modeling our experiments, having a digital twin of the organoid, not of the entire body. Yeah, yeah. Which we can optimize by virtual experiments, doing the real experiments and then going on. So there's really an interesting opportunity here to.
To optimize our experimental approaches. And in general, I think most people are not grasping how much of a technological revolution is taking place at the moment with AI.
[00:32:51] Jack: I think this is all so fascinating and just the types of things that Lulu were just describing, the organoid intelligence, the amount of data that you're able to collect and the potential for you to create these little in silico computerized representations of an organoid might allow you to actually do things to a digital organoid before you do it to the real organoid, before that organoid then gets translated into an animal model and back into the human. Just is like mind blowing to think about that as the potential kind of workflow through which we may be, in the very near future, be able to do drug discovery and perfect our targeting of therapies and whatnot for brains and for cancers and for other things that we really care about. So really bringing computer power to the fore here.
[00:33:42] Tom: I would agree on everything here, and I often have called now for in vivitrosi, which means in vivo, in vitro, and in silico combined.
But I don't see necessarily the sequence of after you showed something in a human relevant system to go to the animal. Because I think that's a step backward.
I think that also what the NIH is now announcing of going to have a human relevant component to in the end confirm the animal experiment, not the other way around, could be actually the way, the way forward. But that's matter of tiny differences in something which says in essence we have to bring all of these ways of getting information together and take the individual advantages.
[00:34:29] Jack: Absolutely. You know, a univariate approach, it seems, you know, simplistic. Right. I mean, you want to know about how all of these things give rise to each other. And I think some of the things that Lulu was referring to are exactly what you want to know. You want to know the temporal sequence. You want to know something changed here, it produced another change in another variable. But you have this organoid system that really allows you to model that not only in, you know, in vitro, but also in silico. And that's a huge, would be a huge step forward. And I'm curious actually, Lulu, are you aware of any kind of special tools or platforms that you've heard about that might lead to this?
[00:35:11] Lulu: Yeah, I think that's very important point about how we really build up those pipelines. Like we have these as a hallmark. So we reach this point, I think we are at the beginning, we try to use the micro array electrode so we can grow the organoid on the electrode so we can monitor the activity of the neurons and then we know what time point they reach, they reach the maturity or they reach the degeneration point or not.
We're still working on that. I think we the technology need to develop to make more sensitive electrode to monitor the activity of the organoid. Also on the other hand, I think everything we work on in the organoid system we need to validate from the postmodern brain tissues, for example, we find a novel target for treatment. Organoid is still a little bit different from the humans. We need to use the human to validate it. That's also one of the important things on the checklist we need to do. So we develop organoid, we generate tons of data, but this data really reflect the human brain or not. We, we needed the human brain data to validate the. So combine these together, we can make sure the organoid of a general routine is not like something just random, just digital twins.
To make sure they are digital-to-digital twins, we need to really compare to the real human brain. The beauty of that is that we use the biopsy, the patient biopsy samples from these live patients, fibroblast cells. We engineer those into the stem cell, the IPSC cells. We use the same patient genetic background to generate the organoid. So when those patients are still alive, still at early stage, we use their individual organoid to test the drug response so we can develop personalized treatment. So in the future, if we can do that, we can make treatment more precise and also help the patients in a more efficient way.
I think that's really important for the future.
[00:37:22] Jack: That's great. Well, to close us out, I want to look ahead and get both of your impressions about what it will take for brain organoids to really become those true platforms for precision medicine. It seems to be the end goal. But what are the biggest hurdles, both in terms of technological challenges, the computational challenges, but also, of course, even the ethical questions that are involved. What do we still have to overcome? Tom, do you want to take that one first?
[00:37:50] Tom: Yeah. I mean, there's a lot in the bucket now from this.
I mean, first of all, the big opportunity using induced permit and stem cells is that these are in essence living biopsies of these people.
So you have some reflection of the individual. And this is very different to the embryonic stem cells which they replaced where the embryo was never born. So you don't know which diseases it would have developed.
And this is a dramatic difference because it is opening up for personalized medicine. But some of the difficulties are also clear. If I take your cells today, some skin or blood cells or whatever, it takes me half a year to get iPC and it gets a year to produce brain organites from them.
So any type of advice I would like to give for you as a person is dramatically delayed.
So this is something we have to overcome. We have to find possibilities to accelerate this, to find opportunities to get to to these faster if you want to build personalized medicine on it. Or we just use them as the training system which where AI helps us to generalize.
The second point, which I think is really important is we have at the moment seeing this dramatic increase in AI technologies that AI is helping science and most people don't understand how fast this is taking place.
AI since 2010, when deep learning was introduced, is doubling its capacities every three months.
So this year's AI systems are eight times more powerful than last year's.
And this is an incredible growth curve.
And combined with all of these technologies to produce big data, we have a completely new period.
If you just think that in the Life Sciences, 3.4 million scientific papers are published every year and only about half of them are open access nowadays.
But still, if you just take these 1.7 million papers which are open access, AI can for two years already annotate these papers better than a human can, can read them in a day, and will never forget.
So we really have suddenly a partner for our research which is integrating all of the knowledge.
And while we are at the moment still looking in how can AI help us?
The question will soon be how can we help AI?
I think that in the future we will write science not with IE, but with AI.
And what I mean by this is that it will tell us if we want to be part of the integrated common knowledge, contribute to what science is about.
We will have to publish open access in machine readable formats. We have to do to close the gaps. We have to listen to AI what is needed to make notes which are informative.
We have to move away from these piecemeal studies where we do an experiment on three substances with four endpoints. But we have to think about the large data sets which really interconnect a lot of things. And you can follow up on these.
So we are really entering with this technological revolution a change in science.
And this will impact on everything. It will also dramatically change how we do our research on brain diseases.
And I see the microphysiological systems as an important pillar. They're producing data human relevant also, possibly on scale, if you only want to.
We can do them on 384 well, plates, we can do whatever and we can feed a lot of information into the AI systems to then say what should be tested in more detail, to prioritize our research.
And this is just scratching on the surface of a vision of science which is solving the big problems like brain diseases with the modern tools of the 21st century, which are these two transformatives technologies.
[00:42:16] Jack: Absolutely. And Lulu, what do you feel are the biggest hurdles that we still have to overcome?
[00:42:23] Lulu: I also agree completely with Tom about make sure those organoids build up cognition, build up memory, and also really have those functions like a human brain with the consciousness. But on the anatomy level, we wanted to form those circuitry like the human brain. Because the organoid we are growing now is just some different cell types. We can mimic particular brain regions. But the human brain is more complicated than that. We have different brain regions, different circuitry in the development and disease, in aging process. All become very different on the genomic level, epigenomic level, transcriptomic level. In the future, we wanted to step-by-step build up the organoid and make them functional with cognition and also recaptulated the anatomy of the human brain. I think that's the hurdle at this moment to, to the organoid brain organoid field.
[00:43:21] Jack: One thing I want to follow up on with this and it sort of gets into an ethical thing just in terms of how we do science now. How especially in the context of like clinical trials where we do animal studies first and then we move them to the human. Do you think that we're ever going to reach a point where the brain derived organoids basically replace animal models which then lead to treatment decisions? Do you think we'll get to that level of precision? And also what sort of data management and analytic methods are we going to need in order to be able to do that? What do you think needs to happen in order for those things to materialize?
[00:43:55] Lulu: The reason why we study organoids, we try to use the human genetic background, but the problem is we can't do behavior test. We can't see the phenotypes, for example the memory loss, cognitive decline or learning disability in this organoid system. That's why sometimes when we find something new from the organoid, we still try to do knockout knockdown in the mouse models to mirror their behavior change to see whether those are the really phenotypes in the patients. For example, the symptoms shown in patients one day. If we can use more precise measurement of the phenotype in the organoid like the cognition, memory, maybe we don't need animal model anymore. So I think linked to what I said last session, I think if we can really evaluate the cognition, the manifestation or symptoms of the organoid, maybe we just get rid of animal models, we can rely on organoids.
[00:44:58] Jack: Tom, what do you think?
[00:44:59] Tom: I mean the first thing is we are not 70 kilogram rats and there's a lot in the sentence and I think that a lot of the lack of progress in a lot of in. In the area of brain diseases comes from the misleading role of some of these animal tests.
It has been estimated that more than 150 billion have been spent on the development of Alzheimer drugs which did not work.
And a lot of this worked in the respective animal models otherwise they would not have progressed.
So I think we need the new players, we need the new things to move us forward. They still have to show that they can do their job.
We have to build trust, we have to optimize them. We have to understand what they can and what they can't. They're not perfect tools.
Nobody should expect this, but they're new kid on the block and it's interesting to see what we can do with these new kids. I believe that people at some point will love about us that we believe that some mice and rats could tell us who live one and a half and two years respectively could tell us anything about neurodegenerative disease which happens after 50 in humans. Yeah, I think the the prospect of really getting to something human relevant is is there and we need to open up for it and need to find the right place for each and every tool we need. If you spend 2.5 billion and more for developing a drug, you cannot say we will not use a certain type sort of information to take the best decision, but we should take the best decision on the rightful place these things have in a development and search strategy.
[00:46:53] Jack: Oh, I agree. And again, a lot of this future is going to be data driven.
It's going to be multimodal. At the moment you mentioned you're collecting a gigabyte worth of data in a particular experiment. I imagine in a few years that will be considered a “cute” amount of data and it will be on the order of terabytes and petabytes with each particular experiment given all the different data types that we'd like to collect. So I look forward to that future.
I think we need to bring this to a close and wrap it up for this episode of Data Points. I really want to thank Lulu and Thomas for their thoughtful and insightful comments into the promise and complexity of this new field of brain organoids. It has been an absolute treat to talk to you both, and as we've heard, this field is really defining rapid innovations, deep interdisciplinary collaborations, and important questions, scientific and ethical alike, all driven by data science and the multiple forms of data which can be collected on these really unique kind of sources of data.
So I guess whether you're a neuroscientist or a data scientist, a bioengineer, just somebody who's fascinated by the future of brain research. We hope that this conversation has sparked some new ideas and raised new questions about the role for data science in the study of brain organoids.
[00:48:17] Monica: Thanks for listening to this episode of Data Points. More information can be found @datascience.virginia.edu. and if you're enjoying UVA Data Points, be sure to give us a rating and review wherever you listen to podcasts. We'll be back soon with another conversation about the world of data science.
