boston—Given that this year's Bio-IT World Conference & Expo includes three tracks on artificial intelligence (AI)-based applications, including a new track focused on Generative AI, this year's conference is It was clear that there would be a huge focus on AI. This was evident from the first keynote address by Dr. Dan Stanzione, executive director of the Texas Advanced Computing Center (TACC) and vice provost for research at the University of Texas at Austin.
In a talk entitled “Unleashing the Power of Advanced Computing in Bioinformatics,” Stanzione said that historically bioinformatics has driven the evolution of TACC, but scientists have been unable to process large amounts of multimodal data. I mentioned that I moved to AI a few years ago as I continued to generate sets.
“Twenty-five years ago you gave people big computers and Fortran and told them to get to work. [but] That’s not the way things are done anymore,” he said. “Life science projects are now done in multiple teams, with each group tasked with executing different aspects of the project.”Modern science requires more than software, he continues. Ta.
Given the size of today's datasets, computing hardware and energy requirements are rapidly increasing. TACC is home to his $60 million supercomputer, Frontera, which was funded by the National Science Foundation and ranks as his 29th most powerful system in the world. TACC also houses Stampede3, a new supercomputer supported by his $10 million grant from the National Science Foundation that advances high-performance computing for AI and machine learning, as well as other platforms. In particular, it began full-scale operation this year.
These and other TACC systems are located in large data centers, consume enough energy to power entire states, and cost millions of dollars to build and maintain, so Ohne joked that it would be better to load computing racks with gold bars instead of boxes.
High power consumption
If current trends continue, electricity demand and costs are expected to rise rapidly in the coming years. In one of his popular sessions at the conference, “Trends in the Trench,” presented by members of the scientific IT consultancy BioTeam, CEO Ari Berman said the AI industry is expected to consume 85% of energy. Ta.–135 terawatts of electricity per year by 2027. By comparison, this is enough energy to power the Netherlands for her entire year.
Such high power consumption is expensive and unsustainable. Companies like NVIDIA have recently announced A new powerful AI computing platform for drug discovery and genomics, which touts the energy efficiency of its graphics processing units. But these are not the only solutions, Berman pointed out, adding that new cooling methods as well as systems that use electricity more efficiently are critically needed.
At several points throughout the conference, a recurring question was asked where AI would be most useful in the life sciences.
Stanzione noted that some stadiums may be less affected. He and conference participants noted that one area where AI can be useful is as a kind of assistant to scientists, helping them synthesize, distill, and extract knowledge from scientific literature. Stanzione also pointed out that the use of AI in personalized medicine may be modest because deep learning algorithms require large amounts of data to be properly trained. However, he added that it could be useful in extracting insights from population-level data, such as on vaccine effectiveness.
Berman called for a more respectful and responsible approach to the application of AI in life sciences. Interest in this technology will not grow unless it is “super useful” and there is a clear use for it, such as pathology or robot-assisted surgery. However, he noted that the most productive use of AI in life sciences is still two to five years away.Berman says there's a lot of hype right now about technology that ultimately isn't well understood.—Also, it's not something you should trust 100%.
discovery of twins
Although not directly about AI, the conference's second keynote focused on digital twins in oncology. This talk was delivered by Caroline Chan, MD, MSc, Vice President and Chief Data Officer and Director of Data Science Development and Implementation at MD Anderson Cancer Center.
In a talk titled “Uncovering Tomorrow's Possibilities: Harnessing the Power of Digital Twins in Cancer Treatment and Research,” Chung discussed both the promise and hype of digital twins and We acknowledged the importance of being thoughtful when implementing technology to ensure this happens. Improve quality of patient care.
Chung joined a multi-agency sponsored initiative led by the National Academies of Sciences, Engineering, and Medicine that focuses on digital twin technology and its applications in healthcare and other fields. Their efforts areGaps and future directions in basic research on digital twins,'', Chung summarized in his keynote speech.
Critical to this initiative's efforts was a clear understanding of what a digital twin is and what it is not. As Chung pointed out, digital twins involve more than simulation and modeling, and the community needs to be “critical about what we define as a digital twin to avoid getting caught up in the hype. “There is.” She also pointed out some of the requirements and challenges in implementing digital twins in the healthcare sector, including the need for transparency, the risk of patient exposure, and the importance of cultural change.
In oncology, digital twins have the potential to be used for clinical trials, drug discovery and development, preclinical testing, and more. Although a digital twin is not an avatar, it “mimics the structure, context, and behavior” of a physical system, and there is a two-way interaction between the virtual and physical counterpart. One of the implications of that, she said, is “an increased risk of personally identifiable information in the data that we're working with.”
Additionally, given that the use of digital twins involves two-way communication, while model validation and validation typically occurs at a single point in time, Chung believes that “continuously changing model validation and validation; and how do we manage the generation of quantification of uncertainty?”
Digital twin developers must address some of the same data challenges as AI developers.
“In today's healthcare settings, data generation and flow is cumbersome, labor-intensive, and standards and terminology vary, resulting in large amounts of duplicate data in electronic health records,” she says. pointed out. There is also a positive publication bias as training data may exclude relevant information important to clinicians that is not currently captured in the EMR, as well as negative outcomes.
To illustrate the data challenges she and her colleagues are grappling with, Chung used the example of tumor imaging, where differences in the protocols used can impact tumor detection and tracking. MD Anderson doctors may use an imaging method to detect a patient's 2 mm tumor. However, that same patient may have visited another center before using a method that could detect 3 mm tumors. Physicians mistakenly enroll patients into clinical trials based on results from one site because they think the tumor is progressing, or enroll patients because they think they no longer need treatment based on results from a second site. It may be possible to cancel it.
Chung highlighted several recommendations from the NAS report, including the need for broader interaction and collaboration to ensure verification, validation, and quantifying uncertainty are key parts of digital twin programs. I ended my keynote speech by emphasizing the points. He also called for careful consideration of the ethical aspects of digital twins before they are used more widely.
Launched in 2002, the Bio-IT World Conference & Expo has attracted an estimated 3,000 attendees from 30 countries, both in person at the event's new home (the Omni Hotel at the Boston Seaport) and online.
