The next frontier in biomedicine: AI models of the human immune system

If the Covid-19 pandemic has shown us one thing, it is how little we really know about how the human immune system works. Despite the remarkable success in development of vaccines to prevent severe disease from SARS-CoV-2, it remains unclear how SARS-CoV-2, a newly emerging virus, causes such a broad spectrum of disease, ranging from asymptomatic and mild cases to severe disease and death. But the reality is that this pattern also holds for other infectious diseases, such as HIV, and for noncommunicable diseases, including responses to cancer immunotherapy, and disease presentation in autoimmune, neurodegenerative, and metabolic diseases as well as allergies.

Decoding and harnessing the power of the human immune system is one of the great frontiers of biomedicine. The immune system consists of an integrated network of genes, proteins, cells, and tissues, billions of times larger and more complex than the human genome. It varies across individuals and over time, due to a broad spectrum of factors including age, genetics, and environment. It is the basis for many of our most important health interventions, from vaccines to emerging immunotherapies.

Artificial intelligence has revolutionized many areas of industry and society and is now providing the tools to unravel the immense complexity of the human immune system and transform the future of human health. A generation ago the human genome was sequenced, at a cost of nearly $3 billion, in a global effort that spanned more than a decade and paved the way for a new era of precision medicine. Today, sequencing a genome requires less than $1,000 and a day. Similarly, the development of single-cell and high-throughput transcriptomics, metabolomics, proteomics, and epigenetics assays, coupled with mass cytometry and biosensors, now allow us to comprehensively assess host responses to infectious and non-communicable diseases, vaccines, and immunotherapies.

Advanced computing and AI tools — including deep learning, probabilistic, and hybrid models, as well as the application of supercomputing resources to conduct advanced simulations of biological systems — can now be harnessed to unravel the complexity of the human immune system. For example, deep learning methods have helped to identify parameters for “immune age,” measures of inflammatory markers that are significantly linked to multimorbidity, immune senescence, frailty, and cardiovascular aging. Moreover, using deep learning, AlphaFold was able to predict the 3D structures of 200 million known proteins.

Developing AI models of the human immune system represents the next revolution in biomedicine. AI is already being applied to identify and improve monoclonal antibodies, and to reveal predictive sequences in immune repertoires relevant to immunotherapy. Imagine a world where we prevent pandemics before they spread, develop effective vaccines in a matter of weeks, and find ways to effectively treat cancer, Alzheimer’s, and other non-communicable diseases.

So how do we get there? We suggest the following guiding principles:

First, this needs to be a global effort. Key stakeholders across public and private sectors must recognize the potential for AI models of the human immune system to mitigate global public health challenges. Such models could help us address the health consequences of climate change; speed up development of universal vaccines to prevent future pandemics; and start to bridge the racial divide in global health. Given sufficient resources and political will, we will be able to spur innovation and lasting social and economic benefits akin to the moon landing, Human Genome Project, or the internet.

Second, we need to develop a roadmap for global coordination, laying out not only a path for scientific discovery, but also a framework for funding, global advocacy, and policy innovation. New global research consortia, with appropriate financial incentives and legal frameworks, will be needed to break down research silos across industry, academia, and government. By better integrating advanced biomedical and computing technologies that reflect the scope of human immunological diversity in the context of large-scale population studies, we will be able to greatly accelerate progress.

Third, new systems will be needed to incentivize data sharing along with standards for privacy, security, access, aggregation, anonymization, and curation. Systems of privacy-preserving machine learning should implement FAIR principles (that is, findability, accessibility, interoperability, and reusability of digital assets) and ensure standardization of the data’s structure and metadata so that these data can more easily be analyzed by AI. Currently, integration and crosstalk between existing databases of human immunology and genetics remains cumbersome at best and non-existent at worst.

Finally, as most of the burden of disease in the world lies in vulnerable populations with biologically distinct immune systems, we need to better understand the biology of humans and immunological diversity. To date, most medical research has been done in healthy adult populations of European descent. To better capture human immunological diversity, research protocols should include those most vulnerable — aging populations, pregnant mothers, newborns, and those living in low- and middle-income countries — while addressing the social and ethical issues of comprehensive immune profiling. Importantly, this research should be done with good clinical practices, including informed consent and in keeping with the highest ethical standards.

We are at an unprecedented moment in history as recent technological advances now offer the unique opportunity to decode the human immune system. AlphaFold’s predictions of protein structures are proof of the incredible power of AI in accelerating scientific discovery. AI models of the human immune system could revolutionize biomedical discovery, transform therapeutic and vaccine development, and enable people to live longer and healthier lives. Will we seize the moment?

Wayne C. Koff is co-founder of the Human Immunome Project. Eric E. Schmidt is co-founder of Schmidt Futures and former CEO of Google. Peter C. Doherty is a Nobel laureate immunologist.