Artificial Intelligence, Big Data, and Regulation of Immunity: Challenges and Opportunities

Studies over the past 80 years have revealed the complexity and the inherently self-regulating capacity of the immune system. While an important aspect of this work has been a better understanding of the regulation of the immune system in health and disease states (Maecker et al. 2012; Kaczorowski et al. 2017; Singh et al. 2019), the interactive and multiple pathways of immunity involving innate and adaptive immune responses have hampered insight and translational research. Immune responses are driven by a network of molecular interactions involving biological response modifiers such as hormones, cytokines, and chemokines, as well as signaling pathways between diverse cells of the immune system. Mechanistically, both antibody-mediated and cellular immunity involve multiple cell types, with or without specific antigen receptors. Immune cells both resident and circulating are thus found in almost all sites and tissues in the body, and can alter homeostatic host responses as well as those that respond to “danger.”

In the context of health, the initiation of immune responses as well as the magnitude and tempo of response is impacted by host genetics, nutritional and environmental factors. Furthermore, immune responses are also deeply impacted by factors such as commensal microbiota and infectious agents (Nikoopour and Singh 2014; Singh et al. 2015). Commensal microorganisms in the gastrointestinal tract, and on mucosal surfaces and skin maintain tissue hemostasis. Pathogens can trigger both innate and adaptive immune responses and cause dysbiosis (Nikoopour and Singh 2014; Singh et al. 2015). The complex relationship between the immune system and microbes needs to be considered in any analysis of regulating immune responses in health and disease. With an evolving focus on “Precision Medicine,” there is also a need to better understand immune regulation for immunotherapy at an individual level (Bluestone and Tang 2015).

The vast and accumulating amount of information about the immune system constitutes big data (BD). The value of BD was stated as “the process of collecting, organizing and analyzing large sets of data to discover patterns and other useful information” (Heymann and Rodier 2004). Existing BD repositories, such as ImmPort, spanning very diverse data sets from clinical studies and trials to microarrays and peptide epitopes, have become important sites for the storage of publicly funded immunology datasets (Bhattacharya et al. 2018; Deng et al. 2022). These repositories contain vast and diverse amounts of information on cellular, molecular, environmental, and microbial interactions that drive innate and adaptive responses, along with the response modifiers, cytokines, and signaling molecules and pathways that regulate immunity and cellular interactions between lymphoid and non-lymphoid cells in all organs and tissues of the body. In the near future, the proposed Human Immunome Project (2023) will generate even more data using “multi-omics” technology over the next few years. The measurements, analyses, and cataloging of immunity BD promises to be fascinating, and challenging, and could be important in generating new information and strategies for regulating immune responses in health and disease (Yu et al. 2019; Chu et al. 2021). While such information and infrastructures are valuable, priorities and processes remain undefined in many cases.

Such an overwhelming amount and diversity of information has prompted an interest in AI, an umbrella term referring to the broad area of computed decision-making. Machine learning (ML), is a subset of AI in which programming enables computer task-learning for making predictions with increasing precision over time. Additionally, “augmented intelligence” relates to the complementary enhancement of human effort via computer systems (Rider et al. 2020). The hope is that these (and future) AI tools will assist us in analyzing BD, thereby further expanding and speeding up the processes of research, diagnostics, and therapeutics in all fields of science and medicine (Bhattacharya et al. 2014, 2018; Deng et al. 2022).

However, immunity-related databases continue to grow, and to be of value this mass of heterogeneous information must be carefully curated. Indeed, the use and accessibility of data are often limited due to the complex structure and lack of uniformity currently in these repositories and platforms. This issue of access must be addressed to make this BD available to the biomedical and clinical research community and to render it analyzable by AI-based tools (Deng et al. 2022). This lack of uniformity will exacerbate as increasingly diverse studies will come to be, such as microbiome research that looks at how molecules produced by microbiota in various body sites interact with lymphoid cells and impact health and disease outcomes. The complex interaction between immune cells and microbiota (Hooper et al. 2012; Nikoopour and Singh 2014; Singh et al. 2015) makes AI/BD analysis more challenging as well as necessary and this has yet to be fully appreciated in regulating immunity (Park et al. 2023).

By merging AI-based analysis with “multi-omics” technology, the next generation(s) of investigators and trainees are presented with an opportunity to advance fundamental research in immunology (Schultze 2015, Wang et al. 2023) and to translate discoveries into clinically relevant outcomes (Bhattacharya et al. 2014, 2018; Chu et al. 2021; Deng et al. 2022). Some recent developments seem promising in this regard, such as repository platforms like ImmPort (Bhattacharya et al. 2018; Deng et al. 2022) that have great potential to advance this process (Fu et al. 2020). A relevant example of improved clinical outcomes from data analysis was in COVID-19 patients with diagnosis and prediction of severity (Liu et al. 2023). We can also look at the rapidly growing field of cancer immunotherapy for promising developments. Significant research efforts are devoted toward finding biomarkers that can assist in disease monitoring and predicting the efficacy of the interventions (Naqa et al. 2018; Jiang et al. 2022; Sharma et al. 2023). The results have also been used to modify treatments although the current success rate of these interventions has been limited (Naqa et al. 2018). The use of chimeric antigen receptor T cells (Sadelain et al. 2017; Hong et al. 2020) and epitope-based precision vaccines (Parvizpour et al. 2020; Yurina and Adianingsih 2022) to target cancer cells are new potentially powerful strategies in cancer immunotherapy. However, the multitude of mutations in tumors represents a huge challenge. AI approaches have yet to address this issue and there remains a significant knowledge gap as to how these therapies target cancer. Hopefully, success rates may improve if Precision Medicine can become a reality. However, AI analysis of BD cannot currently be applied to individual subjects to guide personalized therapy (Garattini et al. 2019).

These few examples represent the tip of the proverbial iceberg. The list of promising avenues is already long, and as AI-based tools develop, this list will likely expand. The area of transplantation could be of particular interest. We might see emerging knowledge that could prevent the need for transplant by modifying the underlying disease state, improving the donor and recipient matching process, identifying organs that can be used that would otherwise be discarded, preventing organ rejection through insights into immunity and drug selection, and even improving postoperative care. Yet, what seems abundantly clear, especially from the few examples in cancer immunotherapy, is that designing effective strategies for immunoregulation will require going beyond data collection and AI pattern recognition. The value of BD and AI-assisted analysis will only materialize if we adhere to one of the pillars of modern science, namely that research strategies must go beyond revealing interesting correlations and move us toward a better understanding of the mechanisms that drive the relevant biological processes (Editorial 2015; Naqa et al. 2018; Sharma et al. 2023; Xie et al. 2023). The ongoing search for a mechanistic understanding of immunoregulation will be furthered by new bioinformatics tools that can be integrated into the design of our research strategies and immunotherapy design (Zhang et al. 2014; Jabbari and Rezaei 2019), but in all these developments, AI should only assist (not replace) experimental research.

As noted above, immunoregulation is inherently extremely complex and involves a host and environmental factors. The analysis of how we naturally respond to pathogens and treat or prevent infection with vaccines or immunotherapies must include a detailed understanding of the role of gut flora in immunity (Jabbari and Rezaei 2019). This is but one of the obvious ramifications that recent research has highlighted. We should not only encourage avenues that will lead to a mechanistic understanding of immunoregulation, but one that foster the discovery of a diversity of pathways and therapies. This means creating databases with sufficiently rich sources of information that include factors occurring at various scales.

Amassing mountains of information is of course one of the outcomes of the newly developing approaches to data analysis. But BD also stands for data, which are of substantial variety, great volume, and accumulates with great velocity. However, while a “faster” rate of accumulation can hopefully translate into faster discoveries and clinical outputs, speed should not be mistaken for progress. Even if autonomous vehicles could travel at 300 km/h on our highways, it does not mean that we should push travel to these limits. Improving knowledge, health outcomes, and saving lives matter of course, but technological development should not be a race to publishing outcomes and producing innovations. Ultimately, if the goal of pursuing AI-based analyses of immunological BD is sustainable flourishing and improved well-being of humanity as a whole, then we must ensure that every step taken remains in line with this overarching good. This might mean intentionally introducing inertia to slow the process, like keeping research and development out of the hands of general intelligence (at least temporarily). Also needed is an enhancement of safety protocols already utilized where we have a more comprehensive assessment and equitable access to innovations to come.

We present some ethical considerations that we think are essential (albeit not exhaustive) as we move forward. We composed these with the assumption that regardless of the means, regulating immunity will always be a formidable problem with unintended consequences, some good and bad ones. We must thus proceed with caution. This implies keeping options open for solutions to address problems as they emerge, maintaining humans as final decision-makers, and moving beyond risk assessment. Running clinical trials is expected to reveal the side effects of interventions and when they are likely to occur. Having this information is crucial as it helps to prevent or mitigate some detrimental situations. Risk assessment allows to manage the situations by taking into consideration the probability of undesirable events. It utilizes classic statistical methods and substantial human inference, whereas AI and ML can automatically classify inputs without much human intervention. This has great power and requires caution, just as entering data into statistical packages without understanding limitations can generate incorrect outputs. Regardless of whether it is AI-assisted or not, the province of risk assessments remains in the realm of what we can know, predict, and hopefully do to reduce the chance and/or impact of these undesirable events. While essential, these measures are insufficient. Science and technology scholars often emphasize that innovations are not neutral, i.e., mere tools that we simply add, remove, and control. They change the world, including us, in substantial and often unpredictable and irreversible ways (Jasanoff 2016). The knowledge, techniques, and medical innovations that will come from using AI and immunological BD should be considered in the same light. Consequently, our ethical evaluations must embrace unpredictability and include broader social and environmental impacts. Table 1 outlines some of the major ethical challenges and possible approaches to address the regulation of immune responses and advancing immunotherapy using AI-based analyses of immunological BD.

In addition to the ethical considerations listed in the previous section, the use of new analytical tools in research and immunotherapy calls for the development of strategic guidelines. The following points will be relevant in this context and Table 2 summarizes the strategies and potential recommendations with the regulation of immunity and enhancing immunotherapy using AI/BD analyses:

The field of immunology has always been driven by conceptual advances that help the discovery research to address substantial and long-standing knowledge gaps. This is evident from the lack of a clear understanding of why cancer immunotherapy has relatively low efficacy for example. It can be extended to other immune system-related knowledge gaps such as why vaccines work or do not work, why the immune system fails to control tumor development, why some transplant patients achieve operational tolerance and do not require drugs while others do not, and why autoimmune diseases or allergies progress despite various immune-regulatory mechanisms that should be operating in vivo. In most cases, this can be attributed to the lack of mechanistic understanding and AI-based analyses may help to design new strategies to modulate immune responses and develop rationale for immunotherapy. However, the collection of more big datasets and chasing AI algorithms is not going to make the field of immune regulation and immunotherapy provide answers as to how our immune system works at a mechanistic or individual level, or what immunotherapies should be considered. AI will not deliver more than it can. While AI may allow us to develop better vaccines, it is unlikely to provide answers to what drives vaccine hesitancy. In summary, AI using BD has the potential to suggest new pathways in regulating immunity and advancing immunotherapy through a better understanding of the basic immunological mechanisms, and to do so in an ethical manner. There is an urgent need to develop and refine how to apply AI/BD-driven ML to advance the field of immunology and immunotherapy in diverse diseases. Despite the pressure to move “faster,” this should be guided by an ethical and rationale framework that will guide the search for new knowledge and develop strategies to maintain health and prevent and cure diseases.