NIH to Inject Healthy Bolus of Data to Sustain the Future of AI for Medical Discoveries

Computational modeling and artificial intelligence (AI) are integrating into medicine and biomedical research at a dizzying pace. Practitioners are utilizing AI in all aspects of healthcare, from the staging of lung cancer nodules on MRI images to workflow management in hospitals.

At the same time, the recording process of physiological measures from smartwatches, smartphones, and sensors in clothes, shoes, contact lenses, and toilet seats—any place where material can rub up against the body—is also rapidly expanding. These burgeoning technologies are incredibly promising but require well-defined data that are freely accessible and understandable by the intertwined web of “intelligent” machines, which reliably and efficiently analyze the data. Researchers can then potentially store, curate, and add these data to a collective knowledge base that will spawn the next remarkable medical advance.

Recognizing both the potential and problems of such an endeavor, the National Institutes of Health (NIH) is funding an innovative approach to get the data right. The NIH Common Fund’s Bridge to Artificial Intelligence (Brige2AI) program seeks to “build the bridge” over which data flows into the biomedical and behavioral research space in a useable form that works for scientists, engineers, physicians, and patients.

“We have a deluge of data, but they are not defined enough for machines to understand,” Bruce Tromberg, Director of the National Institute of Biomedical Imaging and Bioengineering (NIBIB), said. NIBIB is one of the five NIH Institutes and Centers that are leading this enterprise. The Bridge2AI program acknowledges a call from the research community for the generation of new data that are designed for biomedical discovery; such data will propel modern AI/machine learning (ML) models towards scientific discovery, advance a culture of ethical consideration around the data, and create a workforce that is skilled in this new method of scientific data creation.

The Bridge2AI program emphasizes the need to move away from “business as usual” in order to generate data that machines can understand. These data—described as “hypothesis agnostic”—shake free from human-generated hypotheses and inferences. One must possess clear knowledge of the data’s origin, including all aspects that surround data collection (which may comprise multiple streams of measurements and signals from individuals, groups, communities, populations, and regions). The program hopes to instill a culture of ethical inquiry that is based on the grand challenge problems that motivate data collection (see Figure 1). For example, how will we preserve privacy when creating digital twins of ourselves [4, 5]? How should machines handle consent and minimize errors when interpreting genomic data? How do social, cultural, racial, and gender issues affect the collection of wearable sensor data?

The Bridge2AI program acknowledges the fact that truly comprehensive data is only possible through contributions by and data collection from individuals with diverse backgrounds and life experiences. A central goal of the program is to create a sustainable culture change with diversity at the forefront. Such clean, connected, and multi-modal data is then poised to enable AI to find the hidden signals that derive unbiased knowledge and provide insight into human health.

Bridge2AI data generation projects will “stitch together” ethically-sourced data from diverse perspectives that are motivated by biomedical and behavioral grand challenges. And the Bridge2AI Integration, Dissemination, and Evaluation (BRIDGE) Center will ensure that the program’s spirit and culture change goals are infused throughout the life cycle of Bridge2AI-generated data. NIH program managers plan to utilize a nimble funding mechanism that provides increased flexibility for the reconfiguration and recombination of project elements to best meet the Bridge2AI program’s needs. Dissemination of the data, tools, standards, and ethical considerations will include “jamborees” that foster a lively and creative learning environment within the broader scientific community.

The Bridge2AI Working Group hopes that the program will accelerate a culture change that will persist in the long term, even after the program is completed. The Bridge2AI culture change is also reflected in other programs across the broader NIH community, which is committed to the incorporation of ethical considerations throughout the scientific discovery pipeline. For example, the Brain Research Through Advancing Innovative Neurotechnologies (BRAIN) Initiative has been identifying neuroethical issues for consideration. More information is available in the BRAIN Neuroethics report. 

In addition to co-leading the Bridge2AI program, NIBIB serves a critical role in the promotion of novel technologies, including new methods that make these technologies useable and translatable. NIBIB supports the Center for Reproducible Biomedical Modeling, which creates tools for reproducible mechanistic models. NIBIB also leads the Interagency Modeling and Analysis Group (IMAG), which coordinates the Multiscale Modeling Consortium (MSM). In 2019, IMAG and MSM held a pivotal meeting that combined mechanistic modeling with ML. The resulting papers [1, 6] set the stage for the integration of these two modeling modalities to ultimately create digital twins and increase patient safety by reducing medical errors [3]. In 2020, a new working group on Multiscale Modeling and Viral Pandemics formed within the MSM. IMAG and MSM entities also worked to implement “Ten ‘Not So’ Simple Rules for Credible Practice of Modeling and Simulation” [2] on COVID-19 models.

As a bioengineer, Tromberg views mathematics as the cornerstone of the Bridge2AI program. “One of the unifying characteristics of biomedical engineers is our commitment to the idea that biological processes can be represented symbolically, by mathematical equations,” he said in a video for Bridge2AI’s kickoff. “With enough measurements and the right equations, we believe that it’s possible to understand and predict the behavior of any complex biologic system.”

The mathematical and statistical communities that SIAM serves are well positioned to embrace the culture change that is promoted by the NIH Bridge2AI program, the NIH BRAIN Initiative, IMAG, and MSM. We at the NIH encourage readers to establish diverse teams that sustain equitable partnerships and work with data generators to create novel mathematical and statistical methods that address ethical issues surrounding data privacy, research consent, error reduction in scientific and clinical workflows, automatic data, and model annotation. We urge members of the SIAM community to collaborate with mechanistic modelers to design new modeling frameworks that complement AI/ML models. These methods will serve as powerful tools that rapidly accelerate the new future for scientific discovery and improved global health.

References

[1] Alber, M., Buganza Tepole, A., Cannon, W.R., De, S., Dura-Bernal, S., Garikipati, K., ..., Kuhl, E. (2019). Integrating machine learning and multiscale modeling — perspectives, challenges, and opportunities in the biological, biomedical, and behavioral sciences. Digital Med., 2, 115. 
[2] Erdemir, A., Mulugeta, L., Ku, J.P., Drach, A., Horner, M., Morrison, T.M., ..., Myers Jr., J.G. (2020). Credible practice of modeling and simulation in healthcare: Ten rules from a multidisciplinary perspective. J. Translat. Med., 18, 369. 
[3] Makary, M.A., & Daniel, M. (2016). Medical error — the third leading cause of death in the US. BMJ, 353(1), 353:i2139. 
[4] Masison, J., Beezley, J., Mei, Y., Ribeiro, H., Knapp, A.C., Sordo Vieira, L., ..., Laubenbacher, R. (2021). A modular computational framework for medical digital twins. Proc. Nat. Acad. Sci., 118(20), e2024287118.
[5] Niederer, S.A., Sacks, M.S., Girolami, M., & Willcox, K. (2021). Scaling digital twins from the artisanal to the industrial. Nat. Comput. Sci., 1, 313-320. 
[6] Peng, G.C.Y., Alber, M., Buganza Tepole, A., Cannon, W.R., De, S., Dura-Bernal, S., ..., Kuhl, E. (2021). Multiscale modeling meets machine learning: What can we learn? Archiv. Comput. Meth. Engin., 28(1), 1017-1037.