An insider’s view of the effort to decode and model the entire immune system.
We’ve sequenced the genome. We’ve predicted the structures of 200 million proteins, we’ve set out to map the entire mammalian brain, and we came up with vaccines for the coronavirus in record time. So what’s next?
As chair of the board of the Human Immunome Project, I ask you to consider the potential of decoding the human immunome. The impact on science would be immeasurable, and the impact on human health could be even greater.
The human immunome is perhaps the most complex system known to humankind. It’s more than just the networks of molecules and cells we think of as the human immune system. It also includes every single interaction your immune cells, proteins, genes, and tissues have over your entire lifetime, the influence of every infection and every environmental impact, and how all that varies from person to person and place to place. And it’s constantly changing, as are the external elements (environment, pathogens) that interact with it.
So how do you even begin to model all that?
Some of the smartest people I know have said it couldn’t be done, because the science isn’t there yet—and they can’t be blamed for their skepticism. The immune system is unimaginably complex, and unraveling it is going to be way more complicated than sequencing the human genome, mapping the human brain, or accelerating cancer research. But more important than the question of whether we should attempt it is the question of when we should start. And the answer is: Now.
The field of immunology is breaking new ground regularly, aided by systems thinking and computational biology, while the field of artificial intelligence is positively on fire.
Last week, I was in La Jolla, California, for a summit to launch the Human Immunome Project, which aims to build a quantitative and predictive model that will help us better understand these complex interactions and anticipate how an immune system will respond, even if we don’t fully understand how or why. Over the course of three days I listened to big thinkers across multiple fields, from immunology to systems biology to artificial intelligence.
So what’s it going to take?
According to the experts, there are a few major challenges that make this project such a BHAG (Big Hairy Audacious Goal). One is the Tower-of-Babel problem. Researchers are siloed in their still-very-distinct disciplines, where they don’t attend the same conferences and in some instances don’t even really speak the same language. Even when experts do speak to each other, their data may still be siloed. Often there are no incentives to share data—in fact many, many barriers exist that prevent doing so.
“Let’s not let the perfect be the enemy of the good.”
Shai Shen-Orr, a systems immunologist, data scientist, professor at Technion – Israel Institute of Technology, and co-founder of the company CytoReason says the field struggles with a “data insight gap,” where “we generate enormous amounts of data with the latest technologies but are stuck still analyzing it with yesterday’s methods.”
Often, the data doesn’t exist in machine-readable form, and even when it does, it will vary from lab to lab, machine to machine, and researcher to researcher, according to top AI expert Stuart Russell, who led a panel on challenges and opportunities in AI at the summit last week. “Two ventilators from the same manufacturer, the same model number, can produce very different measurements,” he points out as just one example.
The sheer amount of data, of course, looms as another major challenge. We need to collect data from your genome, your epigenome, your proteome, transcriptome, metabolome, lipidome, and microbiome, in addition to codifying your innate immune system as well as your adaptive immune system—and your exposome on top of all that. But lest you be concerned that everything we already know or will discover about the human immunome will surpass the limits of our capacity, it won’t. According to Russell, who literally wrote the most used textbook on AI, “The scale of data to process is not daunting.”
A key scientist on Google’s DeepMind AlphaFold team, Kathryn Tunyasuvunakool, was actually quite encouraging at the summit. There’s lots of data we already have and many people doing interesting work with it, she says. So the question becomes simply defining the right things to ask, collecting the right data to model, and iterating from there. “Let’s not let the perfect be the enemy of the good,” she says.
What impact do we hope to have?
In the next five years, the Human Immunome Project is planning to build a publicly available model of the immune system that will deepen our understanding of biology, speed the development of new drugs, and allow us to get new treatments to people faster.
Where will we be a decade from now? Imagine a personalized model of your immunome, based on your current state of health, your clinical past, and future projections. That would be a powerful tool that could change the course of some disease you might suffer from—or prevent you from getting sick in the first place. It will be able to show us not just correlations, but the underlying causations we can’t even imagine now.
“Immunology goes across all human medicine. Maybe we’ll be changing the world here.”
That’s why I am so enthusiastic about this project. It will materially impact the way we do biology, and help advance machine learning as well. MIT statistician Caroline Uhler put it nicely at last week’s meeting when she said, “Biomedical sciences are uniquely suited not only to being one of the greatest beneficiaries of research in machine learning, but also one of the greatest sources of inspiration for it.”
The organizational efforts of the Human Immunome Project can be galvanizing, by providing frameworks, infrastructure, and incentives for collaboration. Our priority is to structure this in a way that addresses the many data management issues, which include compatibility and integration, security, and intellectual property, not to mention the ethics issues such as patient privacy, informed consent, and inclusivity. But our request for data is not a one-way ask, because the more data we can query, the more robust the models will be, and we will feed those results back out to researchers everywhere, advancing everyone’s work across multiple fields.
By studying and building predictive models of maternal and newborn immunity, we hope to usher in an era of better pregnancies and better outcomes. Modeling how the immune system of the elderly fails, we can potentially slow or combat aging. We expect our model will not only accelerate research across infectious and autoimmune diseases but also boost our efforts to address cancer, heart disease, Alzheimer’s disease, sepsis, and more.
“I’m incredibly excited about the Human Immunome Project, basically because I think we can try to do the kind of research that can really save millions of lives,” said Eric Schmidt, Human Immunome Project honorary conference co-chair, in a video address at the conference.
“Immunology goes across all human medicine,” said Peter Doherty last week, the Nobel Prize-winning immunologist who serves with Schmidt as the honorary co-chair. “Maybe we’ll be changing the world here.”
And you can too. Track our progress at humanimmunomeproject.org.
Author’s note: The views I expressed here are my own, and not necessarily those of the Human Immunome Project.
