Advancing Cancer Research With Large-Scale High Performance Computing
Imagine a powerful new cancer drug that today sits undiscovered, not in a remote jungle, but somewhere among billions of potential molecular structures. Finding it manually is an impossible task.
Yet, researchers are now using some of the world's most powerful supercomputers, capable of performing a mind-boggling quintillion (a billion billion) calculations per second, to hunt for these needles in a molecular haystack. This is the world of exascale computing, and its fusion with artificial intelligence (AI) is revolutionizing our fight against cancer.
This powerful combination is pushing the boundaries of what's possible, turning previously insurmountable challenges into solvable problems. From accelerating drug discovery to decoding the secrets of treatment-resistant cancers, AI and exascale computing are igniting a new era of innovation in oncology, offering new hope for patients and clinicians alike 1 6 .
Finding patterns in massive datasets that humans would miss
Modeling biological systems at unprecedented scale and detail
Accelerating the identification of promising therapeutic candidates
To understand the revolution, we must first understand the tools. Exascale computing represents the next frontier in supercomputing. Simply put, an exascale computer can perform at least one exaflop, or one quintillion calculations, per second.
"It would take every person on Earth performing one calculation per second over four years to match what a machine like Europe's new JUPITER supercomputer can do in a single second." 2
These systems are not just faster; they are fundamentally more powerful, capable of modeling and simulating complex biological systems in ways that were previously unimaginable 1 .
Meanwhile, Artificial Intelligence (AI), particularly machine learning and deep learning, excels at finding patterns in massive, complex datasets. In cancer research, AI can analyze medical images, genomic sequences, and molecular structures to identify subtle signs of disease or promising avenues for new therapies 6 .
Provides the computational power needed for massive simulations and AI model training on unprecedented scales.
Finds patterns in massive datasets that would be impossible for humans to detect manually.
The synergy is clear: AI models require immense computational power to train on large datasets, and exascale computers provide the ultimate engine for this task. As one researcher put it, "The combination of AI plus Aurora allowed us to screen 50 billion small molecules for potential cancer inhibitors in about 20 minutes. On earlier supercomputers, screening even a billion molecules in that amount of time was unthinkable." 1
| Supercomputer Name | Location | Key Capability | Primary Use in Research |
|---|---|---|---|
| Aurora 1 | USA (Argonne Lab) | Exascale Computing | Drug discovery, climate science, materials engineering |
| JUPITER 2 7 | Germany (Jülich) | 1 ExaFLOP/s, ~80 ExaFLOP/s AI performance | Earth system modeling, AI model training, materials science |
| LUMI 2 | Finland | 386 Petaflops (Pre-Exascale) | AI research, climate modeling |
One of the most promising and ambitious applications of this technology is the quest to drug the "undruggable." For decades, certain cancer-causing proteins have resisted all attempts to create chemical treatments against them, earning this daunting label 1 .
A groundbreaking project led by Argonne National Laboratory, in collaboration with the University of Chicago Medicine Comprehensive Cancer Center and funded by ARPA-H, is taking aim at these targets. The IDEAL project focuses on a class of proteins known as intrinsically disordered proteins, which lack a fixed structure, making them notoriously difficult for drugs to bind to 1 .
The research process is a powerful loop between cutting-edge computation and real-world validation.
The process begins with laboratory experiments that identify which specific disordered proteins play a key role in cancer progression 1 .
If the 3D structure of a protein is unknown, researchers use the Advanced Photon Source (APS), a powerful X-ray light source, to determine its shape. The recent upgrade to the APS provides brighter X-ray beams, allowing scientists to capture high-resolution structural details and visualize how potential drugs might bind to the protein 1 .
The protein's structure is then fed into the Aurora exascale supercomputer. Researchers run atomic-level simulations to understand the protein's behavior and identify tiny pockets or crevices where a small molecule could potentially bind and inhibit its function. As one scientist vividly explained, "Imagine a protein being like a piece of chewed-up gum, full of little pockets and crevices. We simulate the structure on Aurora, find those pockets..." 1
With these targets identified, AI tools on Aurora perform a massive virtual screening, sifting through billions of small molecules from digital libraries to find those with the highest potential to bind to the identified pockets 1 .
The most promising candidate molecules from the simulation are then provided to experimental collaborators. They order the physical molecules and test them in the lab, measuring their effectiveness at inhibiting the protein's activity or stopping tumor cell growth 1 .
This integrated approach is already yielding impressive benchmarks. The use of the Aurora supercomputer has dramatically accelerated the initial discovery phase. The ability to screen 50 billion molecules in just 20 minutes provides researchers with a shortlist of highly promising candidates orders of magnitude faster than traditional methods 1 .
Molecules screened in 20 minutes using Aurora exascale computing 1
Proteins now targeted with new computational approaches 1
The scientific importance of this work cannot be overstated. Successfully inhibiting an "undruggable" protein would open up entirely new therapeutic avenues for cancers that currently have poor prognoses. It represents a paradigm shift from accepting certain targets as untreatable to systematically dismantling the barriers that make them so. This methodology lays the groundwork for a more rational, computationally driven drug discovery process that could be applied to a wide range of complex diseases 1 .
| Computing System | Scale of Molecular Screening | Approximate Time |
|---|---|---|
| Traditional Computing | 1 billion molecules | "Unthinkable" to complete in 20 minutes |
| Aurora Exascale + AI | 50 billion molecules | ~20 minutes |
The modern computational cancer researcher relies on a sophisticated digital toolkit. Below are some of the essential "research reagents" and materials that power experiments in this field.
AI algorithms designed to analyze visual imagery, ideal for finding patterns in complex medical scans.
Example: Models like ResNet152 and DenseNet121 analyze mammograms to classify malignant changes with high accuracy (AUC >0.9) 5 .
Scalable deep learning software stacks that allow researchers to build and deploy AI models on supercomputers.
Example: The CANDLE project created a universal platform to tackle key cancer research problems across national lab supercomputers 1 .
A facility that produces intense beams of X-ray light to determine the 3D atomic structure of proteins.
Example: The Advanced Photon Source (APS) is used to visualize the structure of "undruggable" proteins, providing a target for drug design 1 .
The raw genomic data that reveals mutations, gene expression patterns, and other molecular features of cancer cells.
Example: AI platforms integrate this data with clinical records to pinpoint actionable mutations and guide targeted therapy 6 .
The impact of AI and HPC extends far beyond discovering new drugs, touching nearly every aspect of cancer care.
AI algorithms are now being trained to read mammograms, CT scans, and MRIs with accuracy that can match or even exceed expert radiologists. These tools can detect subtle abnormalities earlier, reduce false positives and negatives, and help standardize interpretations. In pathology, AI analyzes digitized tissue slides to distinguish cancer subtypes and reduce diagnostic variability 6 .
The goal of precision medicine is to tailor treatments to an individual patient's unique cancer. AI is vital for sifting through enormous genomic and molecular datasets to identify the specific mutations driving a patient's disease and predict which treatments are most likely to be effective. This helps avoid ineffective therapies and their associated toxicities 6 9 .
The benefits of this technology are being democratized. In Europe, the JUPITER AI Factory (JAIF) will help startups and small businesses access this power 2 . National funding calls, like one in Luxembourg, encourage collaboration between public research institutions and private companies to integrate AI and HPC into their innovation processes 4 .
As we look ahead, experts forecast several exciting trends for 2025. There is a growing focus on moving effective treatments into earlier stages of disease and on developing the next generation of therapies, such as allogeneic "off-the-shelf" cell therapies and more sophisticated cancer vaccines 9 .
The ability of AI to analyze hematoxylin and eosin (H&E) slides to predict treatment response and resistance is another area of intense interest 9 .
The convergence of artificial intelligence and exascale computing is more than just a technical achievement; it is a fundamental shift in our approach to understanding and treating cancer.
By combining the pattern-finding prowess of AI with the unparalleled simulation power of exascale systems, researchers are solving problems that were once considered intractable.
From screening billions of drug candidates in minutes to taking aim at previously "undruggable" cancer targets, this powerful synergy is accelerating the entire journey from scientific discovery to patient impact. While challenges remain, the fusion of AI and exascale computing is undoubtedly lighting the path toward a future where cancer can be detected earlier, treated more effectively, and ultimately, defeated.