NIH’s Bridge to Artificial Intelligence Facilitates Scientific Collaboration

September 26, 2022 - Bridge to Artificial Intelligence (Bridge2AI) is an NIH Common Fund Program focused on combining machine learning and artificial intelligence to provide advancements in the biomedical industry. According to a market analysis report done by Grand View Research, between 2019 and 2021, the AI market in healthcare increased by 167.1%. Using research, data, and analytical tools, this program will help facilitate scientific collaboration and biomedical progress. Paul Boutros, PhD, MBA, professor in Human Genetics and Urology at UCLA, sat down with PharmaNewsIntelligence to explain the program and his role in it.

An Overview of Bridge2AI

While giving an overview of the project, Boutros shared that “Bridge2AI is an attempt by the NIH to set standards for how AI can be used in healthcare. It's a template, a way to push the field forward and show demonstrations of what could be done. It's not meant to cover every tool or data type, but a few areas that serve as role models or templates.”

“One of the most important things about Bridge to AI is that it's not a project led by one person or one type of person. It's not led just by computational modeling people like me, cardiac clinicians, oncologists, or translational researchers. It's led by a group that includes every aspect of clinical care and translational medicine. And that's why it has a chance to impact the world,” he emphasized.

Boutros mentioned that “the data is being sourced from a huge multimodal range of places. Some come directly from medical records, and some are information generated during clinical care but are more holistic from a research perspective.”

Bridge2AI Goals

Although the goals of this project are widespread, Boutros explained the program was developed with healthcare challenges and their solutions in mind. Beyond just optimizing tools, the program’s purpose is to optimize tools for general use. Paul reiterated, “my team is not looking for tools that will work for the 0.1% of elite academic medical centers. We're looking for tools that have generalizability. That means generalizability across healthcare systems and the ancestral, socioeconomic, and other diversities of the country and the world.”

Boutros highlighted the way his team optimized and benchmarked tools. “My team doesn’t just look at beautiful clean data generated by perfect research or a clinical trial in a top-tier academic medical center. Instead, they look at a range of data sets coming from places with different characteristics, noise, and amounts of missing data. Even the data generation projects embody that, where they include urban and rural centers.”

Tool Optimization

Boutros’s role is predominantly focused on optimizing the tools used in healthcare. He shared that within the project structure, there are multiple groups, including data generation groups and his group, which focuses on tool optimization.

“What my group is dealing with are all the tools that go from taking data — generated either directly as part of clinical care or data that's generated as part of research projects — and integrating those and performing AI on them,” stated Boutros.

Challenges in Tool Optimization

In his discussion with PharmaNewsIntelligence, Boutros explained several challenges the team anticipates in this project.

“For the tool optimization core, it's going to be important to find the balance. It is important that the team doesn’t focus only on data quality and what the data is used for. In that case, my team would have good data, but we miss things that could change the world and improve healthcare. Additionally, you can't jump into sophisticated artificial intelligence machine learning models with bad data and then find garbage,” explained Boutros.

To address this concern, Boutros stresses that “the data versus modeling challenge, or the balancing, is incredibly tough. Bridge2AI is putting together an external advisory board to help give us guidance on it. The program works closely with all our colleagues in the different cores to figure out the core milestones and guidance of success they care about. That will help us balance that.”

“Ultimately, the optimization team is surveying all the tools used in the projects across all Bridge2AI. The team is going to evaluate their maturity. Which ones have been used by hundreds of thousands of people, and which ones have been used by one group and maybe need a lot of optimizations or testing? They might be great, but maybe healthcare needs to benchmark and validate that to be confident in them,” he cautioned.

Additionally, Boutros mentioned that he finds there are two main problems in tooling in algorithms. The first of the two is not knowing the proper or best strategy.

“The other one is what if there's scarcity? What do researchers do if there are only one or two tools? So, the first thing to do is figure out why. Is it because there's not enough data, or is it because people can't get access to the data so that they can't figure out how to work with it? That's a scenario where the tool core can be engaged with other people and as resources to bring in new algorithm developers,” he remarked.

Impacts of the Program

While the impacts of this program cannot be measured until years down the line, Boutros speculated on some of the potential benefits of this Bridge2AI. Specifically, he explained to PharmaNewsIntelligence the potential public health benefits and collaborative successes.

Public Health

The public health impacts of this project are multifaceted. Boutros points out that “anytime tools are improved, it has big impacts on every aspect of medicine. Every aspect of biomedicine and basic research is impacted. It allows us to do things faster, easier, with less expense, and with increased accuracy.”

According to a survey conducted by Healthcare Information and Management Systems Society, 42% of healthcare professionals working in IT, business, and clinical practice believe that AI holds the most promise when it comes to infectious diseases.

PharmaNewsIntelligence asked Boutros how he thinks these tools may benefit future pandemics. “Some research projects, the data generating projects, might be related to whatever comes next. That's possible. But the core concept of Bridge2AI is finding those core tools that will lift all boats and enrich the biomedical workforce’s skills in ways that will enable it to be more capable and more efficient when the next pandemic strikes,” he responded.

Scientific Collaboration

In addition to the potential public health benefits, Boutros proposes how this project will facilitate scientific collaboration.

This initiative will work toward making research more collaborative in multiple ways. “First, a core idea of the Bridge to AI is that there will be core data sets that'll become standards that everybody uses. Those data sets will be attached to tools that everybody can access and use — this is open data and open-source software here. That's not the same thing as collaboration, but it certainly means that it reduces the silos, where one group's data or innovation or software doesn't help anybody else,” indicated Boutros.

Beyond the universal access to data and tools, this program is also taking steps toward scientific collaboration by seeking advice from multiple other organizations. “One of the key things Bridge2AI is doing within tool optimization is taking these new tools and benchmarking with our colleagues,” asserted Boutros.

He described how, using these tools, his team will develop a gold standard data set to be shared globally. Boutros shares that his team will ask scientists worldwide how they would analyze and approach ideal and noisy data sets.

Part of the goals of this project is to “bring together a community of people around the tools and methods that are interested in the optimization of what we're doing.”

“Collaboration in science is always tricky because people think about it as collaboration and competition. But the hope is that, between the open source, open data, crowd-based benchmarking strategies, and the approaches of bringing people together sequentially as the project evolves and researchers reach areas of interest, we'll be able to take steps at least to mitigate that.”