Episode 3 Part 1 - Genomics Futures: AI and Synthetic Biology

Show notes

Speakers:

• Mathew Garnett, Group Leader also at Wellcome;
• Alexandra Canet, science communicator and producer of the Genomics Futures podcasts;
• Muzlifah Haniffa, Deputy Director at the Wellcome Sanger Institute;
• Alta Charo, emeritus professor of law and bioethics at the University of Wisconsin;
• Robert Smith, lecturer of Science and Technology Studies at the University of Edinburgh.
• Episode description:

This third episode is split into two parts. They are both looking at two workshops that were titled Understanding and engineering cells and genomes for the future and Innovative technologies for measuring and engineering life. We merged the participants, themes and reflections in one same episode as most of the themes overlapped and reflections were quite similar.

In part one, we spoke with workshop organisers, Professor Muzzlifah Haniffa and with Dr Mathew Garnett, who introduced us to some of the themes that arose during the workshops, such as synthetic biology and the use of Artificial Intelligence for the advancement of genomics. We then spoke with Dr Alta Charo and with Professor Robert Smith to look at the ethical side of the questions posed by the future of genomics.

Transcript

Matthew Garnett 00:00
So people have been proposing doing large scale genome synthesis for nearly, well, actively proposing for 10 years, it’s a much longer field of research. But the question has always been, ‘why’ and I would love to have a more concrete conversation about, well, what actually are we going to do with this kind of technology.

Alexandra Canet 00:22
What will the future of genomics look like in 2050?

Matthew Garnett 00:26
And I think when we look forward 25 years from now, I can imagine that we are beginning to be able to create artificial systems that approximate a tissue, the complexity of a tissue, and potentially an organ.

Alexandra Canet 00:43
Welcome to the third episode of the Genomics Futures podcasts. This is part one of a two part episode in which we’ll explore two of the workshops, one that looked at the future of understanding and engineering genomes and cells, and another that was titled ‘Innovative Technologies for Measuring and Engineering Life’. The themes and visions that came out of both workshops were complementary and similar, so we interviewed participants from both workshops together. I’m Alexandra Canet, Science Communicator and producer of these podcasts, and we’ll be taking you through our conversations with attendees throughout part one and part two of this episode. Today, in part one, we will first hear from workshop organisers, Professor Muzz Haniffa, Deputy Director at the Wellcome Sanger Institute, and Dr Matthew Garnett, Group Leader, also from the Wellcome Sanger Institute. They each chaired a different workshop and were interviewed together for this podcast. We asked them about their visions for 2050. We start with Muzz.

Muzz Haniffa 01:49
I like to think of Genomics in the context or concept of language of DNA, and what we have will be individual letters that we can construct to make a word, a word that makes a sentence, a sentence that makes a paragraph, and a paragraph that makes a document. And at this point in time, we know the letters, they’re just four, unlike the English alphabet, but how those letters come together and how they then create meaning and context is not completely understood. So we have the letters that make up the cells in the body, but how that then gets translated into the words and context that make up the cellular function, we still do not understand and are exploring, and eventually we can use the language and construct, from scratch using those letters, how we can make cells that don’t necessarily exist, naturally, tissues that don’t necessarily exist, or enhance tissues, enhance cells. The opportunity is limitless.

Alexander Canet 03:00
Fabulous, fantastic. Matthew, what is your vision?

Matthew Garnett 03:05
So, Alex, one of the kind of ways we approached this in our workshop was to think about who would be a user of genomics in the future, and as a kind of archetype individual, we elected to select a student, probably in a lower- or middle-income country, who had a question or a biological question, or maybe even wanted to design an experiment, and they could effectively query some type of user-friendly large language model, AI-type interface and run in silico or computational experiment that would then produce a result for them and potentially, ultimately lead to the design or execution of a new experiment. And I think there’s a lot of things that were kind of implicit in this right one was, I think, importantly, democratisation of genomics, which we think is really important. The other is obviously having that knowledge bank that you can draw upon, and that’s gonna to take the generation of very substantial data sets that allows us to build up accurate, predictive computational models that people can use, as well as those kind of interfaces that allow someone to actually query those databases. Now that’s just one example, but then you can imagine extending it into, you know, the clinical setting, for example, as Muzz was elaborating on, where you could, you know, use some of those kind of capabilities to query around, for example, a particular patient and leverage that bank of knowledge to help guide that particular patient’s treatment.

Alexandra Canet 04:59
Okay. Fantastic. And what were the biggest themes coming out of the workshops?

Muzz Haniffa 05:05
I’ll keep to the language analogy that I used at the start. There were many issues that came out, and the biggest challenges is, how do we actually understand the language? And how do we then create new content from the language? So this was the understanding biology, which is the reading and understanding the language, and generative or synthetic biology, using genomics, which is, how do we then create new content? And that has many key issues, keeping still to the language analogy. How would the understanding of the world be if we were only focused on books written by a group of individuals in Europe and in North America without actually reading the literature that’s produced around the world? And this will have significant impact of understanding biology in a global setting, and likewise, that would significantly impact what we then produce in terms of how that is going to be valuable for global benefit, and not just in a few. You can’t just rely on a small group of individuals and genomics information contributed by the few to benefit the global community.

Matthew Garnett 06:29
And Muzz you, you talked about kind of generative genomics and inducing like different cell states, maybe ones that aren’t observed naturally, kind of physiological, that’s a human, and I think most people can understand, you know, how you would use genomics to personalise medicine, but I think I’d be interested to know why is it going to be important to be able to engineer synthetic states that may not actually naturally occur?

Muzz Haniffa 07:01
Yeah. I mean, I think if you- if we actually look across the human lifespan, for example, there are some interesting insights from when we develop and a lot of the cells are quite different and they have different functions. But if you look at the adult state, you no longer see those cells. So sometimes we forget what nature actually makes. But there is the space of what may not actually exist. Even though it’s biologically plausible, it’s all about the route to get to the cell state. So one of the emerging themes from that workshop was about creating synthetic blood, and if we can actually find a way of making, for example, a red blood cell, which is extremely efficient at transporting oxygen, but could last much longer and has a much quicker way for us to make it and will be adapted to the person that needs it. These are the things that we could potentially synthesise, and it may have very different properties to the red blood cell as we know it, but essentially, it performs the function of a cell that can deliver and transport oxygen, which each of us need to sustain life.

Alexandra Canet 08:15
And Matthew, what about your workshop? What were the themes and the main key takeaways?

Matthew Garnett 08:21
Yeah, in our workshop, there was more of a technology focus. And actually there were, there were a lot of themes that emerged, but I think there were two that we really ultimately ended up focusing on. And one was the ability to write DNA efficiently. And at the moment, we can synthesise small pieces of DNA, fragments of DNA, and we can stitch them together but actually it would be highly desirable to be able to synthesise or design much larger segments of DNA, and that would allow us to interrogate the design properties of genomes, to model disease much more faithfully than we currently can, and ultimately, think much more about designing synthetic states or highly efficient genomes that may be more suited for manufacturing purposes, even, for example, manufacturing a red blood cell. Now, some of the barriers that we identified is it, it’s still, you know, expensive to do that, and one of the at least key challenges is that it’s still difficult to get large pieces of DNA into cells, at least mammalian cells and human cells. And so some of the areas for development there about driving down cost, but also improving that technology. There are also some ethical issues that come from that around the ability to efficiently write DNA, and what would be the appropriate usage of those. The second major theme was really about how artificial intelligence is going to be disrupt to the way we do science and genomics going forward.

And to be honest, the group struggled, because it just felt like there were so many potential points at which it may impact so, let me give you an example. The most obvious is in our ability to interpret large data sets and find valuable information in a way that a human brain currently can’t, but then there was also this kind of idea that actually it’s not a one way direction. There’s this idea that actually AIs could design experiments. AIs could even design experiments potentially with human input, but then could effectively commission experiments through robotic platforms that would actually generate the data to either support or refute a particular hypothesis or to fill in the gaps in the knowledge that would improve on that model. I think that bit may even go beyond the next 25 years, but there was a sense that things like automation, robotics and the ability of computers to talk to those platforms could be really very transformative in the way we actually do science.

Muzz Haniffa 11:38
But coming back to the two points you raised, Matthew, in terms of being able to write larger segments of DNA and also being able to use AI, one of the questions we addressed before was, how do we make this representative, and how do we make this equitable? How do you see in those two areas where we can actually achieve representation and diversity and equity?

Matthew Garnett 12:10
Yeah, yeah, I know it’s a really important point. So in terms of our thinking around writing genomes, I suppose if we could do it efficiently enough, we would be in a position to explore all possible diversity effectively, both represented in human populations, but as we discussed, potentially diversity in new code or new words that allow us to develop new products, new cell types that don’t exist in nature. So I think there is a lot of scope to be able to do that. I think there are, there are definitely challenges. One is that the underlying dataset has to represent that diversity. Otherwise, we will train models that are not fit for purpose across, you know, different populations. So I think there’s a risk there. There’s also a risk in making that data a public resource. If the data sits inside commercial silos, then that may limit the wider access to that, right? So if those large scale data sets that effectively train the largest AI models become commercial products, there is a potential risk that those are not widely accessible. And then a last point being, how do you make that accessible outside highly modern societies, but in societies where access to resource, it’s much more difficult to access those technologies, even to access those AI models.

Alexandra Canet 13:43
Another theme that I saw that came up from both workshops was the vision to create human proxies and tissues in a dish. But I think it’d be good to reflect on where we are now, what is possible now, and what do you want to achieve in 25 years time?

Matthew Garnett 14:00
Yeah, absolutely. So the first thing to say is, having experimental proxies is very important. What do we mean by that? We mean that we have model systems in the laboratory that we can test hypotheses on, for example, interrogate the function of a gene that represents a human state or condition, cell or tissue or organ. And the reason that is important is it allows you to draw a direct connection between an experimental change and the effect that you’re interested in. So it allows you to derive causality. Now in my own personal area of expertise, which is the biology of cancer, we’re getting pretty good at growing cells in the laboratory. So these are cancer cells that are derived from human tumours.

And we can do this with a reasonably high efficiency and in such a way that we can sample a large amount of what we call heterogeneity or diversity within both different types of cancers and the different types of cells that exist within those cancers. However, one of the challenges we face, even when we work with cells, is we don’t always have the right type of conditions in the laboratory that most faithfully mimic actually what’s happening, for example, with a human, where you have the additional complexity of the vasculature, the stroma, so things like fibroblasts or the immune system, all of which impact on the way cells function. So a real challenge for us is really understanding what are those types of environments that exist, and how can we recapitulate them in the lab?

I think when we look forward 25 years from now, I could imagine that we are beginning to be able to create artificial systems that approximate a tissue, the complexity of a tissue and potentially an organ. I do think, even in 25 years, actually recapitulating in an experimental system, the complexity of an organ will still be very challenging for us.

Alexandra Canet 16:22
Muzz, what about your thoughts?

Muzz Haniffa 16:25
So let’s imagine the human body, and let’s imagine taking cells, which are the building blocks of the human body, and trying to keep that alive in a dish. So there are lots of technical challenges, obviously, but what we have now is the fact that we can grow many different types of cells, not just by taking cells and tissue from the human body, but also from stem cells that you can make from a terminally differentiated cell. So you can take a skin cell and then transform it into a stem cell, and then you can make a brain cell, you can make a skin cell again, and you can make a muscle cell, etc. So that’s been a radical advancement. And now we can actually put these cells together and see how they come together, or what you call cellular ecosystems, and that’s the start of the tissue, and you can then start to see how they then form organs, and how they can then form an assembly of interconnected organs. Right now, we have some good models of what we call organoids, i.e. cultures, that recapitulate organs. But they may not be complete in the sense that they don’t have the blood vessels, it is not perfused, it is not being oxygenated in the same way that the cells in the body would be. It may not have the complete armamentarium of the immune cells that would be surveying the human body. So there are some shortcomings, but we’re getting there, with engineering. What I also think would be very exciting to consider is how we can use this knowledge to perhaps enable regeneration inside the human body. So instead of transferring cells, we actually provide the stimulus and the right supporting factors that will allow your liver to regenerate when it’s been damaged, potentially your ligaments to regenerate. There are lots of wears and pairs that you know we could actually help with. So lots of regenerative potential. And one striking area where this is going to make a massive impact is the brain. If we could ensure that the brain can regenerate, that would be fantastic. So that’s one area where you can have regeneration inside the human body. But I’d like to take the future all the way where this may not necessarily have to be done in a dish, that this may actually be virtual. And we can then create things and create digital avatars of our cells, our tissues, our organs, ourselves, in essence, and could be potentially extremely transformative in terms of what it could predict with regards to response to treatment, projection of disease evolution and how we will age and our lifespan.

Alexandra Canet 19:39
We also spoke to Dr Alta Charo, Emeritus Professor of Law and Bioethics at the University of Wisconsin, and Dr Robert Smith, Lecturer of Science and Technology Studies at the University of Edinburgh, who gave us their own perspectives on the topics to hand, artificial intelligence, synthetic biology and new genomics technologies.

Alta Charo 19:58
When I think about the challenges, I divide them between the technological challenges and the social challenges. And on the social end, I think we’re facing two very big ones. The first is the business model for genomics, at least in the area of therapeutics. That business model that we’ve had successfully for 50 years in developing pharmaceuticals just doesn’t fit. Pharmaceuticals will earn you money at a penny a day for every pill that people take for the rest of their lives. But with genomics, what we’re looking for are one time, one and done cures for which the opening price is the only price, the only chance to earn a profit. So the price tag comes in at millions. It’s unsustainable, and without some system for amortising those costs over the lifetime of a patient and the full population, we are not going to find a way for companies to successfully bring these incredible technologies to patients. The second challenge is equitable access, because some of these technologies necessarily not only require a great deal of money, but also a great deal of expertise and specialised equipment. And while some diseases are widespread and may be amenable to this kind of approach, we have 1000s and 1000s of rare diseases that require bespoke therapies, and in both these cases, whether it’s economic or numerical orphan status, we don’t have a system that can adequately assure that the technologies really benefit everybody over the long term.

Alexandra Canet 21:36
Fantastic. Thank you, Alta. Rob, what are your thoughts?

Robert Smith 21:41
My take is similar, but slightly different and more meta, perhaps. I think that we’re reaching a point where when you’re talking about genomics, that the shift that we’re going through is from reading frame to a writing frame. So being able to write large genomes from scratch, being able to synthesise ever large pieces of DNA and use those in many different ways. So being able to put them towards de-extinction programmes, to use them to develop new gene therapies, to use them to produce new products. These are all things that are becoming possible, I think, with genomics. And for me, the challenge is we’re in a way, trapped in the way that we think about what we can do with these technologies, but we’re also trapped in the way that we can think about distributing benefit and equity and addressing questions of fairness and things. So actually, what Alta was talking about with business models is one of the ways in which we’re trapped by existing ways of thinking. So for me, it’s this, this is the one of the main challenges.

Alta Charo 22:52
Listening to Rob’s point about the move from reading to writing reminds me of another kind of overarching observation looked at very simplistically, one might think about the public’s reaction to science in the 19th century, at least in the Western world, as one of celebrating the inventive power of science and engineering. And then into the 20th century, with World War One and chemical warfare and then nuclear power, nuclear explosions, we find a fear of science because of its destructive power. So from its inventive to its destructive and at the end of the 20th century and into now our 21st I think the public’s concerns have shifted again toward the creative power of science, not what it can destroy, but a fear of what it can create, whether it was IVF babies in the 80s, or it’s the fear about the use of heritable genome editing, or even now the debates around de-extinction. And I think this is a kind of underlying angst that needs to be understood and addressed if we’re going to really take advantage of these technologies in a way that benefits people, while not throwing up their concerns about not just untoward environmental effects, but about exceeding what ought to be limits on human control over themselves and their environment.

Alexandra Canet 24:17
So artificial intelligence is increasingly dominant in genomics research, and it needs vast amounts of data. What do you think as a society we will need to put in place to make sure that the use of data is ethical, but at the same time, it sustains innovation in the genomics space?

Robert Smith 24:34
I don’t think that there is one answer to this question, and I don’t think that there is a single societal response. I think it’s been very clear from some of the recent debates that there are actually quite different cultural positions on how we should treat data, on what the rights of individuals are to that data, and how that should be settled legally. I mean, a really obvious example is just to look at a comparison between the US and EU approaches but you can go much further.

I also think that one of the challenges with the way that we talk about AI is in a universalist way, as though, if we just feed all of the data to a computer, it’s going to magically produce all of the right answers that we need for genomics. But when you talk to practitioners in this space, it becomes very clear that there are lots of small data problems within genomics. A lot of the data sets that you need are actually highly specialised. They’re often actually very expensive to produce. And so one of the things that is gonna happen, I think, in the next 10 years, is actually a set of very specialised data sets being produced which are specifically for AI training and AI driven genomic discovery work. And if you take that as the thing which is happening, you end up in a different place when you start talking about, how do you ensure the benefits are equitably distributed. About the use of the data is ethical, that it is adequately protected, because then those data sets are much smaller and much more specific, and then the kinds of people that you’re talking about become concrete, and the kinds of problems also become concrete.

Alta Charo 26:38
My thinking about this has changed over time. I used to be somebody who was very comfortable with the idea that if we had an adequate social compact in which the benefits of data use and analysis flowed to all the people from whom data had been collected, then widespread collection would be justified, and I was less concerned about the dignitary harms or invasions of privacy that lacked any concrete consequence, whether to your life insurance or your health insurance or your social stigmas, things like that, because I thought the value of the data outweighed those kinds of concerns, and that was a position I held for a while, until I began watching one country after another begin to topple toward a more authoritarian and more surveillance based state that really threatens civil liberties. So I’m now much more inclined to be focused on ways in which you can ensure the anonymization of data, analysis at a population scale, rather than an individual scale, or even a small cell size scale, so that our ability to maintain a zone of privacy around ourselves, one that is free of government observation and intrusion, as well as private sector invasion and intrusion has become more and more important to me. Perhaps more a political issue than an ethical issue. I’m not sure what the answer is, because I’m not enough of a technologist, but I look at things like blockchain technology with great hopefulness about ways that you can create linkages that preserve the usability and the value of data and data analysis, while also creating obstacles to having it penetrated to the level of being able to identify an individual for a particular purpose. But again, I can’t speak to the technologies that might make it possible, and I do wonder if AI might be helpful in the development of these kinds of technologies, whether it’s through genuine random numbers, you know, generation, or something else that might help us get there.

Alexandra Canet 28:50
Yeah, absolutely. You both attended different workshops, but what did you feel was missing from the conversation?

Alta Charo 29:01
One of the challenges in genome editing in particular, has been differentiating between near term realistic applications and long term realistic applications and science fiction fantasies or dystopias. For me, some of the near term applications that are getting less attention than they deserve have to do with the ability to scale up what are clearly bespoke therapies into something that can be more widely made available. We have a recent success story with sickle cell disease in the United States with an ex vivo genome editing approach that is simply unsustainable, not only because of the cost and the specialty facilities, but because it requires a bone marrow conditioning regimen that is incredibly arduous, as well as requiring hospitalisation for extended periods. And how to scale that kind of thing up is something that is a near-term and very real issue that deserves a lot of our attention and goes directly to issues around business and around equitable access, and yet a conversation will quickly spin into discussions around designer babies, as if people who can already have children the old fashioned way are going to rush out for difficult, burdensome, expensive procedures to get some marginal purported benefit having to do with editing of embryos, which is not even technologically feasible yet. I guess it’s not so much that it was missing, but perhaps an absence of emphasis on real problems that we need to be solving immediately, as opposed to wondering about things that are far off in the future.

Robert Smith 31:00
Yeah, we agree here, which is boring. I think that this is really hard to talk about futures. It takes a lot of work, and so often those things get folded together, right? Like just in the kind of normal day to day way that we make sense of futures, or use the kind of tropes that we have our disposal and and the slightly disorienting thing that often happens in these kinds of workshops is you feel like you’re getting thrown between, like, very concrete technical advances and then these, like, wild, yeah, like Jurassic Park or designer baby kind of chats that are going on. I think I really appreciate the space that scientists have to think about what is possible and to kind of explore the edge of the possible and, and to push the boundaries there. But I think too often the conversation starts from the technology and not from the problems. And so actually, if you started with this question of like, well, how do we advance healthcare in a way which is equitable, which means it needs to be cheap, easy to distribute, has a set of criteria associated with it, and then what that’s the future that we want to pursue? And then what place does science have within that? Like, how do we pursue scientific or technological trajectories, begin to deliver on that. I think you would end up in a slightly different place when you have that kind of conversation.

Alexandra Canet 32:32
Throughout this conversation, equity has come out quite a bit, and it was also something that came out very much during the workshops. But I’d like to ask you, what does equity mean in this context?

Robert Smith 32:47
So I guess the classic definition of equal versus equitable is, equal is giving everyone the same thing and equitable is giving people the thing which is necessary and according to their means, and our workshop was slightly diverse in terms of the people that were at the table, but I think it could have been much more so, and so the first step in any of these kinds of conversations, when we talk about equity, is to make sure that the people that we think may be inequitably treated at the moment are present in the room and given voices and given the power to imagine the futures.

Alta Charo 33:33
So for me, I think it’s important to distinguish between equitable access or equitable outcomes in the context of medical care, as opposed to human health, because if we talk about human health, then achieving something remotely equitable requires moving on from medical therapies and focusing on the other areas of genomic application, environmental sensors that help to identify sources of tainted water, and genomic editing that helps create more nutritious and resilient crops that can feed people at a large scale without destroying their local environments. These are the kinds of advances that would actually promote human health far more than any number of individual medical therapies. We want to think about therapeutics, the narrower question about access to healthcare, then, I think we’re looking at a combination of things that involve technological innovations that make things easier to produce or to deploy in under-resourced settings. So it means moving from ex vivo to in vivo kinds of approaches. It means moving to regulatory systems that are based on approving platforms, as opposed to approving each therapy individually. It involves moving to delivery systems that can be managed in non refrigerated, non specialised facilities with infusions that are easily managed or therapies that in the future are even orally taken. These are the kinds of things where technology can actually overcome some of the social obstacles that prevent us from extending the reach of these therapies. But even with that, I don’t think it begins to touch on the human health inequities that really need a much more fundamental kind of approach.

Alexandra Canet 35:50
Thank you for listening to the third episode of the Genomics Futures podcasts. As mentioned in the introduction, this is part one of episode three, which has two parts. If you haven’t listened to the first two episodes of this series yet, you can find them wherever you get your podcasts, under Genomics Futures.

Alexandra Canet 36:09
Part two of this episode will continue looking at the themes evolving around artificial intelligence and synthetic biology with Ben Lehner from the Wellcome Sanger Institute and the Centre of Genomic Regulation in Barcelona. Kondwani Jambo from the Liverpool School of Tropical Medicine in Malawi, and Patrick Boyle, a specialist in automation and digital data for biology. If you want to get in touch, please do. You might agree, disagree, or have your own thoughts about the topics and themes discussed in these conversations. We’d love to hear them. You’ll get in touch with us at genomicsfutures@sanger.ac.uk