%63%65%73
-
KAIST Uncovers the “Hidden Key” to Immunotherapy for Intractable Brain Tumors
Researchers have uncovered a clue to why immune checkpoint inhibitors—cancer therapies that release the immune “brakes” exploited by tumors to evade attack—show limited efficacy in some brain tumors. A KAIST research team found that B cell and antibody responses initiated in tumor-draining lymph nodes, rather than T cells alone, are critical to the antitumor effects of anti-CTLA-4 therapy, opening a new avenue for treating intractable brain tumors.
KAIST (President Choongsik Bae) announced on the 19th of July that a research team led by Professor Heung Kyu Lee from the Department of Biological Sciences has identified a previously unrecognized immune mechanism through which anti-CTLA-4, a type of immune checkpoint inhibitor, promotes B-cell responses in tumor-draining lymph nodes, thereby helping the immune system attack brain tumors.
Glioblastoma is one of the most aggressive malignant brain tumors, with frequent recurrence and a poor prognosis even after surgery and radiation therapy. Immune checkpoint inhibitors, which restore the ability of immune cells to attack cancer cells, have produced substantial therapeutic benefits in various cancers. However, their effectiveness in glioblastoma has remained limited because of the highly immunosuppressive environment surrounding the tumor.
Researchers have traditionally regarded T cells—immune cells that can directly attack cancer cells—as the primary target of immune checkpoint inhibitors. B cells, meanwhile, are well known for producing antibodies following infection or vaccination, but their role in brain tumor immunotherapy has remained largely unexplored. The research team therefore investigated whether anti-CTLA-4 could influence B-cell responses as well as T-cell responses.
The findings challenged the prevailing T-cell-centered view. In mouse glioma models, anti-CTLA-4 treatment reduced tumor burden and significantly prolonged survival. However, these therapeutic effects were largely lost in mice lacking B cells, demonstrating that B cells are required for the efficacy of anti-CTLA-4 treatment in these models.
The team also identified where B cells played their key role. Rather than being prominent in the brain, where the tumor cells were located, the response increased markedly in the Deep Cervical Lymph Nodes, which are located deep in the neck and receive lymphatic drainage from the brain. In particular, germinal center B cells and T follicular helper cells, both of which are important for antibody formation, increased together in these lymph nodes. This was accompanied by an increase in immunoglobulin G, or IgG, responses. IgG is a major class of antibody that can recognize cancer cells as targets and help immune cells eliminate them.
The resulting IgG antibodies bound to the surface of glioma cells, helping macrophages—immune cells that engulf foreign substances and cancer cells—remove the tumor cells more effectively.
The research team also examined this process directly in vivo. Using a specialized dual-reporter glioma model expressing the red fluorescent protein mCherry and the green fluorescent protein EGFP, the researchers successfully visualized tumor-infiltrating phagocytes actively engulfing glioma cells following anti-CTLA-4 treatment.
The study provides the first functional evidence that B-cell immune responses—previously known mainly for their roles in infection and vaccination—can be a key factor determining the effectiveness of immunotherapy for hard-to-treat brain tumors. It also expands the conventional T-cell-centered framework of cancer immunotherapy by showing that treatment efficacy can be strongly shaped by immune responses originating not only within the tumor, but also in tumor-draining lymph nodes outside it.
Yumin Kim, a postdoctoral researcher in the KAIST Department of Biological Sciences, served as the first author of the study, with Professor Heung Kyu Lee serving as the corresponding author. Professor Ji Eun Oh from the KAIST Graduate School of Medical Science and Engineering also contributed to the research. The findings were published on July 10 in Science Immunology.
Paper title: The efficacy of immunotherapy in glioma requires distal B-cell responses in tumor-draining lymph nodes
DOI: 10.1126/sciimmunol.adz2494
Authors: Yumin Kim, In Kang, Byeong Hoon Kang, Won Hyung Park, Chae Won Kim, Hyun-Jin Kim, Jeongwoo La, Myoung Seung Kwon, Sang Hee Park, Seo Hyeon Im, Hyeon Cheol Kim, Keun Bon Ku, Minji Kim, and Ji Eun Oh from KAIST, with Heung Kyu Lee from KAIST as the corresponding author.
This research was supported by National Research Foundation of Korea grants RS-2023-NR077244 (to H.K.L.), RS-2024-00439735 (to H.K.L.), RS-2024-00411928 (to Y.K.), RS-2025-00517107 (to J.E.O.), and RS-2026-25509011 (to H.K.L.). This study was also supported by Samsung Science and Technology Foundation grants SSTF-BA1902-05 (to H.K.L.) and SSTF-BA2201-11 (to J.E.O.).
2026.07.20 View 100 -
A Breakthrough Concept in Nano-Printing Technology
<(From Left) Professor Inkyu Park, Ph.D candidate Byung-Ho Kang, Researcher Jun-Ho Jeong, Professor Junseong Ahn>
A breakthrough technology has been developed that allows metal circuits floating on water to be transferred directly onto any desired surface. A South Korean research team has introduced a novel technique capable of transferring ultra-fine nano-circuits onto plant leaves and fruits, as well as curved automotive surfaces and robot exteriors, all without causing any damage. This technology is expected to be widely utilized across various cutting-edge industries, including smart agriculture, wearable healthcare, and bioelectronics.
KAIST announced on June 15th that a joint research team led by Distinguished Professor Inkyu Park from the Department of Mechanical Engineering, Dr. Jun-Ho Jeong from the Korea Institute of Machinery and Materials and Professor Junseong Ahn from Korea University has successfully developed "Water-Floating Nano-Transfer Printing (WF-nTP)." This technology enables precision metal thin films floated on water to be transferred onto various 3D surfaces.
Conventional nano-transfer printing (nTP), which is widely used to manufacture electronic devices and sensors, typically requires high heat, intense pressure, adhesives, or toxic chemical solvents. Consequently, applying this method to biological tissues or complex curved surfaces that are sensitive to heat and pressure has proven highly challenging.
To overcome these limitations, the research team proposed a completely new approach: floating metal circuits on the surface of water.
Fabrication: The team deposited an ultra-thin layer of metals such as gold (Au), platinum (Pt), palladium (Pd), or nickel (Ni) onto a polymer mold.
Separation: They then used plasma gas which has a high-energy state of ionized gas to selectively remove (etch) a part of the mold.
Floating: When this structure is placed in water, water infiltrates through the microscopic gaps, causing the 20-nanometer (nm) thick metal film to float to the surface on its own while perfectly maintaining its original shape.
<Overview of the water-floated nano mesh and versatile transfer technique>
The research team transferred the metal circuits using a "scooping" method, which involves submerging the target object beneath the floating film and slowly lifting it upward. As the water evaporates, capillary force (the force that moves liquid through narrow spaces) tightly adheres the circuit to the surface. Once the water completely dries, intermolecular forces come into play, securing the circuit firmly in place without the need for any adhesive.
Notably, the team also succeeded in transferring circuits onto hydrophobic (water-repellent) surfaces, such as lotus leaves. By adding a small amount of ethanol to the water to lower its surface tension (the property of a liquid surface that causes it to shrink), they effectively overcame a major limitation of conventional technologies.
This technology holds immense potential for widespread application because it can adapt to diverse surfaces while flawlessly preserving the nano-patterns. Using this method, the research team fabricated Surface-Enhanced Raman Scattering (SERS) sensors which are used for high-sensitivity detection of trace chemical substances and attached them directly to plant leaves and fruit surfaces. Through this, they successfully detected thiram, a pesticide component, on the surfaces of leaves and fruits. Furthermore, they successfully implemented a wearable, high-performance hydrogen gas sensor by transferring a palladium (Pd) mesh onto highly flexible thermoplastic polyurethane (TPU) fibers.
<Research Image(AI-generated)>
Distinguished Professor Inkyu Park stated: "This technology is highly significant as it shatters the substrate limitations of conventional nano-transfer printing, allowing nano-patterns to be transferred onto sensitive surfaces like living plant leaves or human skin without heat or adhesives. We expect it to evolve into a core platform technology for wearable sensors and bioelectronics, finding applications in smart agriculture for pesticide detection without damaging crops, wearable health monitoring devices, bioelectronic devices, and electronic skins for next-generation robots."
This research, with PhD student Byung-Ho Kang from the KAIST Department of Mechanical Engineering participating as the first author, was published online on March 30, 2026, in the prestigious international academic journal Nature Communications.
The study authors include Byung-Ho Kang, Ji-Hwan Ha, Yeongjae Kwon, Sohee Jeon, Donho Lee, Byeongmin Kang, Soon Hyoung Hwang, Junseong Ahn, Jun-Ho Jeong, and Inkyu Park.
Meanwhile, this research was conducted with funding from the Ministry of Science and ICT through the National Research Foundation of Korea (NRF) Mid-Career Researcher Support Program, and the Ministry of Trade, Industry and Energy through the Korea Evaluation Institute of Industrial Technology (KEIT) Alchemist Project.
2026.06.15 View 1635 -
Longevity mediated by suppressing age-associated circRNA
< (Back row from left) Prof. Yoon Ki Kim, Prof. Seung-Jae V. Lee, and Gwangrog Lee; (Front row from left) Dr. Sung Ho Boo, Sieun S. Kim, Seokjin Ham, and (top) Donghun Lee >
Cells in our bodies produce RNA based on genetic information stored in DNA, and RNA serves as a blueprint for making proteins. Researchers at our university have discovered a new phenomenon: removing 'circular RNA' that accumulates in cells as we age can slow down aging and extend lifespan. This study provides crucial clues for uncovering the principles of aging and developing treatment strategies for related diseases.
Professor Seung-Jae V. Lee’s research team (RNA-Mediated Healthspan and Longevity Research Center) from the Department of Biological Sciences, in collaboration with research teams led by Professors Yoon Ki Kim and Gwangrog Lee, announced on the 18th that they discovered the RNASEK protein—an enzyme that degrades circular RNA—plays a vital role in slowing aging and extending lifespan.
Until now, circular RNA has been regarded mainly as an aging marker because of its stability, which allows it to accumulate over time. However, the molecular mechanism for removing this RNA and its direct link to aging had not been clearly identified. The research team conducted this study to determine how the accumulation of circular RNA affects aging and whether an intracellular management system exists to regulate it.
Using Caenorhabditis elegans (C. elegans), a short-lived roundworm widely used in aging research, the team first confirmed that the circular RNA-degrading enzyme RNASEK is essential for longevity. They also discovered that as aging progresses, the amount of RNASEK decreases, resulting in an abnormal accumulation of circular RNA within cells.
Conversely, artificially increasing the levels of RNASEK (overexpression) extended the lifespan and allowed the organisms to survive longer in a healthy state. This implies that the process of appropriately removing cellular circular RNA is critical for maintaining health and longevity.
The research team also found that RNASEK prevents the toxic aggregation of circular RNAs in aged organisms. . When RNASEK is deficient and circular RNA accumulates, "stress granules" form abnormally inside the cell, which can impair cellular functions and accelerate aging.
RNASEK works alongside the chaperone protein HSP90 (which helps proteins avoid misfolding or clumping) to inhibit the formation of these stress granules and help cells maintain a normal state. Notably, this phenomenon was observed not only in C. elegans but also in human cells. In mammals, RNASEK also functions to directly degrade circular RNA; a deficiency of RNASEK in human cells and mouse models led to premature aging.
< Diagram showing progress toward longevity or aging depending on circular RNA and the removal enzyme RNASEK >
The researchers explained that this study is significant as it identifies a mechanism for regulating aging at the RNA level. They suggested that research using RNASEK to control circular RNA could lead to the development of treatment strategies for human aging and degenerative diseases.
Professor Seung-Jae V. Lee of KAIST, who led the study, explained, "Until now, circular RNA was merely regarded as a marker of aging that accumulates over time due to its stability. This study proves that circular RNA accumulated during aging actually induces aging, and that RNASEK, which removes it, is a key regulator that slows aging and induces healthy longevity."
< (AI-generated image) Longevity induced by the circular RNA-removing enzyme RNASEK >
Drs. Sieun S. Kim, Seokjin Ham, Sung Ho Boo, and Donghun Lee from the KAIST Department of Biological Sciences participated as joint first authors. The research results were published on February 24 in the world-renowned scientific journal Molecular Cell.
Paper Title: Ribonuclease $\kappa$ promotes longevity by preventing age-associated accumulation of circular RNA in stress granules
DOI: 10.1016/j.molcel.2026.01.031
This research was conducted with support from the Leader Researcher Program of the National Research Foundation of Korea.
2026.03.18 View 2774 -
KAIST Develops Brain-Like AI… Thinks One More Time Even When Predictions Are Wrong
<(From left) Professor Sang Wan Lee, Myoung Hoon Ha, and Dr. Yoondo Sung>
Artificial intelligence now plays Go, paints pictures, and even converses like a human. However, there remains a decisive difference: AI requires far more electricity than the human brain to operate. Scientists have long asked the question, “How can the brain learn so intelligently using so little energy?” KAIST researchers have moved one step closer to the answer.
KAIST (President Kwang Hyung Lee) announced on the 29th that a research team led by Distinguished Professor Sang Wan Lee of the Department of Brain and Cognitive Sciences has developed a new technology that applies the learning principles of the human brain to deep learning, enabling stable training even in deep artificial intelligence models.
Our brain does not passively receive the world. Instead of merely perceiving what is happening in the present, it first predicts what will happen next and, when reality differs from that prediction, adjusts itself to reduce the difference (i.e., prediction error). This is similar to anticipating an opponent’s next move in Go and changing strategy if the prediction turns out to be wrong. This mode of information processing is known as “Predictive Coding.”
< Predictive Coding (PC) Module >
Scientists have attempted to apply this principle to AI, but encountered difficulties. As neural networks become deeper, errors tend to concentrate in specific layers or vanish altogether, repeatedly leading to performance degradation.
The research team mathematically identified the cause of this problem and proposed a new solution. The key idea is simple: instead of predicting only the final outcome, the AI is designed to also predict how its prediction errors will change in the future. The team refers to this as “Meta Prediction.” In simple terms, it is an AI that “thinks once more about its mistakes.” When this method was applied, learning proceeded stably in deep neural networks without halting.
<Analysis of Instability in Predictive Coding Model Errors>
The experimental results were also impressive. In 29 out of 30 experiments, the proposed method achieved higher accuracy than the current standard AI training method, backpropagation. Backpropagation is the representative learning method in which AI “goes backward by the amount of error and corrects it.”
Conventional AI training methods (backpropagation) require tightly interconnected layers, meaning the entire network must be computed and updated simultaneously. In contrast, this new approach demonstrates that, like the brain, large AI models can be effectively trained even when learning occurs in a distributed and partially independent manner.
<Performance Comparison of Predictive Coding Models>
This technology is expected to expand into various fields where power efficiency is critical, including neuromorphic computing, robot AI that must adapt to changing environments, and edge AI operating within devices.
Distinguished Professor Sang Wan Lee stated, “The key to this research is not simply imitating the structure of the brain, but enabling AI to follow the brain’s learning principles themselves,” adding, “We have opened the possibility of artificial intelligence that learns efficiently like the brain.”
This study was conducted with Dr. Myoung Hoon Ha as the first author and Professor Sang Wan Lee as the corresponding author. The paper was accepted to the International Conference on Learning Representations (ICLR 2026) and was published online on January 26.
※ Paper title: “Stable and Scalable Deep Predictive Coding Networks with Meta Prediction Errors”Original paper: https://openreview.net/forum?id=kE5jJUHl9i¬eId=e6T5T9cYqO
This research was supported by the Ministry of Science and ICT and the Institute of Information & Communications Technology Planning & Evaluation (IITP) through the Digital Global Research Support Program (joint research with Microsoft Research), the Samsung Electronics SAIT NPRC Program, and the SW Star Lab Program.
2026.03.06 View 2675 -
KAIST Extends Its Deepest Condolences on the Passing of the Late Chairman Chang Sun Jung, Founder of Jungheung Group
KAIST extends its deepest condolences on the passing of the late Chairman Chang Sun Jung, founder of Jungheung Group.
Chairman Jung made significant contributions to the development of Korea’s construction industry and regional economy, and was a visionary leader who deeply recognized and actively supported the importance of nurturing science and technology talent. In particular, through his generous contribution to the KAIST Development Fund, he left a meaningful legacy in fostering future scientific talent and advancing research environments that will shape the nation’s future.
KAIST honors Chairman Jung’s noble spirit of giving and dedication, and will continue to strive to ensure that his vision lives on through the advancement of science and technology in Korea.
We extend our sincere condolences to the bereaved family and to the executives and employees of Jungheung Group, and pray for the eternal rest of the deceased.
2026.02.03 View 1421 -
Discovery of a Switch to Halt Adipocyte Generation
< (From left) Dr. Ju-Gyeong Kang, Ph.D candidate TaeJun Seol, Professor Dae-Sik Lim >
Metabolic diseases such as obesity, fatty liver, and insulin resistance are rapidly increasing worldwide, but fundamental methods to regulate the process of fat formation remain limited. In particular, once adipocytes (fat cells) are formed, they are difficult to reduce, making treatment challenging. Amidst this, a research team from our university has discovered the existence of a ‘switch’ that prevents fat formation. This discovery elucidates how an ‘epigenetic switch’—which regulates gene activity without altering the DNA sequence itself—functions during the process of adipogenesis, presenting new possibilities for the precise control of obesity and metabolic diseases in the future.
The research team, led by Professor Dae-Sik Lim and Professor Ju-Gyeong Kang from KAIST’s Department of Biological Sciences, announced on January 25th that they have identified ‘YAP/TAZ,’ key regulators of the Hippo signaling pathway*, as playing the role of an ‘epigenetic differentiation inhibition switch’ during the process of adipocyte differentiation**. The team proposed a new mechanism in which YAP/TAZ extensively inhibits the activation of genes responsible for adipocyte formation through its downstream target, ‘VGLL3.’ *Hippo signaling pathway: A cellular control system that regulates when cells grow, stop dividing, and differentiate. **Adipocyte differentiation: The process by which preadipocytes (or stem cells) transform into mature adipocytes.
Cell differentiation is not a simple matter of a single gene turning on or off; it is a complex, organic process involving multiple genes and DNA regulatory regions. The research team tracked the entire process of preadipocytes* differentiating into adipocytes using Next-Generation Sequencing (NGS), which allows for the simultaneous analysis of gene expression changes and epigenetic modifications. *Preadipocyte: A developing intermediate-stage cell whose direction as to which cell it will become has already been determined.
As a result, they confirmed that under conditions where YAP/TAZ is activated, the genetic program that establishes adipocyte identity fails to operate, and the overall adipocyte differentiation network—centered around PPARγ*—is suppressed. *PPARγ: The ‘metabolic master switch’ regulator that controls energy storage and utilization in the body.
Specifically, through single-cell analysis of adipose tissue, the research team identified VGLL3 as a novel target gene of YAP/TAZ. While it was previously known that YAP/TAZ directly binds to and inhibits PPARγ, this study revealed that VGLL3 indirectly controls the entire adipocyte differentiation program by suppressing ‘enhancers,’ which are the DNA regulatory regions of adipocyte genes. This signifies that the Hippo signaling pathway plays a crucial role in regulating the core timing that determines when and how robustly fat cells are created.
Dysfunction of adipose tissue is deeply linked to various metabolic diseases such as obesity, insulin resistance, and fatty liver. The research team expects that further studies on how the YAP/TAZ–VGLL3–PPARγ axis regulatory principle involves adipocyte formation and functional abnormalities will provide new clues for regulating or treating metabolic diseases.
< Schematic Diagram of Adipocyte Gene Regulation >
Professor Dae-Sik Lim stated, “This study is the first to establish that adipocyte differentiation is precisely controlled at the epigenetic level, beyond simple gene regulation. It has laid an important foundation for a more sophisticated understanding of the mechanisms behind adipocyte identity changes and, in the long term, for developing personalized treatment strategies for patients with metabolic diseases.”
This research, with Ph.D. student TaeJun Seol and Dr. Ju-Gyeong Kang as co-first authors, was published on January 14th in the world-renowned international academic journal, Science Advances. ※ Paper Title: YAP/TAZ-VGLL3 governs adipocyte fate via epigenetic reprogramming of PPARγ and its target enhancers, DOI: 10.1126/sciadv.aea7235
Meanwhile, this research was conducted with support from the Leader Researcher Support Program and the Overseas Excellent Scientist Recruitment Program of the National Research Foundation of Korea, funded by the Ministry of Science and ICT.
2026.01.26 View 2374 -
KAIST to Showcase K-Tech Competitiveness at KAIST Pavilion during CES 2026
< Figure 1. Bird's-eye view of the KAIST Pavilion at CES 2026 >
KAIST announced on January 2nd that it will participate in the Consumer Electronics Show (CES 2026), held from January 6 to 9, 2026, at Eureka Park in the Venetian Expo, Las Vegas. KAIST will operate a dedicated 111㎡ "KAIST Pavilion" to showcase its innovative technologies to global corporations and investors.
A total of 12 startups will participate in the KAIST Pavilion at CES 2026. Notably, eight of these companies are built on core AI technologies, highlighting KAIST's artificial intelligence research capabilities and its achievements in technology commercialization.
The participating companies will unveil solutions targeting the global market across various high-tech sectors centered on AI, including robotics, bio-health, hardware devices, and content technology.
One of the standout participants is Hypergram, a faculty-led startup. Hypergram will introduce 'HG VNIR Pro,' the world’s first commercialized compressive hyperspectral imaging technology. This product is an end-to-end solution that integrates industrial-grade precision hardware with AI analysis software, capable of detecting minute chemical properties invisible to the human eye in real-time. During the exhibition, the company plans to demonstrate high-precision AI image analysis using its high-speed hyperspectral machine vision camera.
< Figure 2. Hypergram >
MOSS, a winner of the CES 2026 Innovation Award, will exhibit an AI-based, all-in-one mobile music production platform. This platform allows the general public and indie musicians to create high-quality music without a heavy financial burden. Visitors can experience the intuitiveness and innovation of the technology at an AI experience zone, where they can create music by linking the MOSS app with the dedicated hardware, 'MOSS Pocket Studio.'
< Figure 3. MOSS (Innovation Award Winning Product) >
BareulEye is developing a medical AI diagnostic solution that detects high-resolution changes in internal organ microstructures based on AI-powered quantitative ultrasound analysis technology. At CES 2026, they will unveil a 3D volume reconstruction technology that utilizes smart mirror-based self-ultrasound imaging.
Leveraging this technological prowess, BareulEye recently secured approximately $10 million (14 billion KRW) in a strategic Series A investment from a leading global healthcare company. Led by Professor Hyeon-min Bae, the Director of the KAIST Institute for Startup & Entrepreneurship and CEO of BareulEye, the company plans to accelerate joint technology development and overseas market expansion.
< Figure 4. BareulEye >
In addition to these, various KAIST startups leading innovation in AI, bio, and hardware devices will participate to solidify their foundations for entering the global market.
The KAIST Pavilion is designed with an open layout and large-scale LED displays to attract attention, featuring independent spaces for each company to facilitate in-depth technical presentations and investment consultations.
"Through CES 2026, we aim to imprint the AI-driven innovative technologies of KAIST startups on the global stage and establish a practical bridgehead for their international expansion," said Keon Jae Lee, Vice President of the KAIST Institute for Technology Value Creation.
2026.01.02 View 4250 -
Mathematicians Solve Cellular Noise, a Long-standing Challenge in Biology
< (From left) Researcher Dongju Lim, Researcher Seokhwan Moon, Professor Jae Kyoung Kim (KAIST), Professor Jinsu Kim (POSTECH), Professor Byung-Kwan Cho (KAIST) >
Why does cancer sometimes recur even after successful treatment, or why do some bacteria survive despite the use of powerful antibiotics? One of the key culprits identified is "Biological Noise"—random fluctuations occurring inside cells. Even when cells share the same genes, the amount of protein varies in each, creating "outliers" that evade drug treatments and survive. Until now, scientists could only control the average values of cell populations; controlling the irregular variability of individual cells remained a long-standing challenge.
A joint research team—led by Professor Jae Kyoung Kim (Department of Mathematical Sciences, KAIST), Professor Jinsu Kim (Department of Mathematics, POSTECH), and Professor Byung-Kwan Cho (Graduate School of Engineering Biology, KAIST)—has theoretically established a "Noise Control Principle." Through mathematical modeling, they have found a way to eliminate biological noise and precisely govern cellular destiny. This achievement in securing precision control technology at the single-cell level is expected to be a new milestone in solving challenges in cancer treatment and synthetic biology.
While cells in our bodies strive to maintain homeostasis for survival, their internal environments are constantly changing. Existing genetic circuit technologies could regulate the average protein levels of a cell population but often ended up amplifying the "noise"—the variance between individual cells. The research team compared this to a "shower that fluctuates between boiling and freezing." Even if the average water temperature is set to 40°C, a normal shower is impossible if the water alternates between scalding and icy. Similarly, a small number of cells that escape control due to this "trap of the average" become the primary cause of cancer recurrence or antibiotic resistance. To solve this, the team devised a new mathematical model called the "Noise Controller (NC)."
The researchers first investigated whether they could control the variance of outputs—which differs from cell to cell—using a "dimerization reaction," where the final products of a system bind together to form pairs. In the process, they confirmed that the dimerization reaction could act as a sensor to detect fluctuations (noise) in the cellular state. However, initial attempts showed that this method alone had limits in reducing differences between cells. Consequently, they determined that a device was needed to immediately reduce substances if they were overproduced. They combined this with a "degradation-based actuation" principle, which promptly breaks down proteins when they become excessive. As a result, they theoretically implemented "Noise Robust Perfect Adaptation (Noise RPA)," which maintains a constant noise level despite external environmental changes. Through this, they succeeded in suppressing cell-to-cell deviation to a Fano factor of 1—the minimum level achievable by universal biological systems.
< Figure 1. Conceptual Diagram of Noise Controller (NC) Effects: When no control technology is used (top, gray), the average value of the cell population changes due to external stimuli. With existing control technology (middle, blue), the average value is maintained, but the deviation between individual cells (noise) remains large. In contrast, using the Noise Controller (bottom, green) maintains the average while also reducing the noise level of individual cells. >
The research team proved the model's performance by virtually applying it to the DNA repair system of E. coli. In the existing system, the amount of DNA-repairing proteins varied so greatly between cells that approximately 20% of the cells failed to repair and died. However, by applying the Noise Controller (NC) to unify protein levels across all cells, the mortality rate was slashed to 7%. The team significantly boosted cell survival rates through sophisticated mathematical principles alone. This is highly significant as it moves beyond the "average control" paradigm to realize "single-cell control," dealing with each cell with precision.
< Figure 2. Structure of the Noise Controller (NC).In the conventional control scheme (left), the final output (X2) produces one of the controller proteins (Z2), and this protein is degraded together with the other controller protein (Z1) that generates the system input (X1).In contrast, the noise controller (NC) established in this study (right) has a largely similar structure, but is characterized by the production of the controller protein (Z4) through a dimerization reaction of the final output. This protein directly degrades the system input (X1).Through this mechanism, mathematical expressions for the mean of the final output (lower left equation) and its noise (lower right equation) can be derived >
Professor Jae Kyoung Kim, who led the research, stated, "The significance lies in bringing cellular noise—which was previously dismissed as luck or coincidence in biological phenomena—into the realm of controllable factors through mathematical design." He added, "It will play a vital role in fields requiring precise cellular control, such as overcoming cancer treatment resistance and developing high-efficiency smart microorganisms." Co-corresponding author Professor Jinsu Kim of POSTECH emphasized, "This research demonstrates the power of mathematical modeling, starting from theoretical formulas of intracellular noise using reaction network theory and leading to the design of actual biological mechanisms."
< Figure 3. Actual Biological Circuit Structure of the Noise Controller (NC): A representation of the mathematical model established by the research team implemented as a genetic circuit, which is an actual biological system. The existing control technology (left) consists of a reaction where the final product produces an anti-sigma factor (RsiW), which then binds with the sigma factor (SigW) that generates the system’s input value. The Noise Controller (NC) (right) similarly utilizes the binding reaction between an anti-sigma factor (RseA) and a sigma factor (ECF); however, the primary differences are that the anti-sigma factor (RseA) is produced through the dimerization reaction of the final product , and that the anti-sigma factor (RseA) directly degrades the system’s input value >
The results of this study were published on December 24 in the international academic journal Nature Communications (IF=15.7).
2025.12.29 View 2997 -
How Does AI Think? KAIST Achieves First Visualization of the Internal Structure Behind AI Decision-Making
<(From Left) Ph.D candidate Daehee Kwon, Ph.D candidate Sehyun lee, Professor Jaesik Choi>
Although deep learning–based image recognition technology is rapidly advancing, it still remains difficult to clearly explain the criteria AI uses internally to observe and judge images. In particular, technologies that analyze how large-scale models combine various concepts (e.g., cat ears, car wheels) to reach a conclusion have long been recognized as a major unsolved challenge.
KAIST (President Kwang Hyung Lee) announced on the 26th of November that Professor Jaesik Choi’s research team at the Kim Jaechul Graduate School of AI has developed a new explainable AI (XAI) technology that visualizes the concept-formation process inside a model at the level of circuits, enabling humans to understand the basis on which AI makes decisions.
The study is evaluated as a significant step forward that allows researchers to structurally examine “how AI thinks.”
Inside deep learning models, there exist basic computational units called neurons, which function similarly to those in the human brain. Neurons detect small features within an image—such as the shape of an ear, a specific color, or an outline—and compute a value (signal) that is transmitted to the next layer.
In contrast, a circuit refers to a structure in which multiple neurons are connected to jointly recognize a single meaning (concept). For example, to recognize the concept of cat ear, neurons detecting outline shapes, neurons detecting triangular forms, and neurons detecting fur-color patterns must activate in sequence, forming a functional unit (circuit).
Up until now, most explanation techniques have taken a neuron-centric approach based on the idea that “a specific neuron detects a specific concept.” However, in reality, deep learning models form concepts through cooperative circuit structures involving many neurons. Based on this observation, the KAIST research team proposed a technique that expands the unit of concept representation from “neuron → circuit.”
The research team’s newly developed technology, Granular Concept Circuits (GCC), is a novel method that analyzes and visualizes how an image-classification model internally forms concepts at the circuit level.
GCC automatically traces circuits by computing Neuron Sensitivity and Semantic Flow. Neuron Sensitivity indicates how strongly a neuron responds to a particular feature, while Semantic Flow measures how strongly that feature is passed on to the next concept. Using these metrics, the system can visualize, step-by-step, how basic features such as color and texture are assembled into higher-level concepts.
The team conducted experiments in which specific circuits were temporarily disabled (ablation). As a result, when the circuit responsible for a concept was deactivated, the AI’s predictions actually changed.
In other words, the experiment directly demonstrated that the corresponding circuit indeed performs the function of recognizing that concept.
This study is regarded as the first to reveal, at a fine-grained circuit level, the actual structural process by which concepts are formed inside complex deep learning models. Through this, the research suggests practical applicability across the entire explainable AI (XAI) domain—including strengthening transparency in AI decision-making, analyzing the causes of misclassification, detecting bias, improving model debugging and architecture, and enhancing safety and accountability.
The research team stated, “This technology shows the concept structures that AI forms internally in a way that humans can understand,” adding that “this study provides a scientific starting point for researching how AI thinks.”
Professor Jaesik Choi emphasized, “Unlike previous approaches that simplified complex models for explanation, this is the first approach to precisely interpret the model’s interior at the level of fine-grained circuits,” and added, “We demonstrated that the concepts learned by AI can be automatically traced and visualized.”
< Overview of the Conceptual Circuit Proposed by the Research Team >
This study, with Ph.D. candidates Dahee Kwon and Sehyun Lee from KAIST Kim Jaechul Graduate School of AI as co–first authors, was presented on October 21 at the International Conference on Computer Vision (ICCV).
Paper title: Granular Concept Circuits: Toward a Fine-Grained Circuit Discovery for Concept Representations
Paper link: https://openaccess.thecvf.com/content/ICCV2025/papers/Kwon_Granular_Concept_Circuits_Toward_a_Fine-Grained_Circuit_Discovery_for_Concept_ICCV_2025_paper.pdf
This research was supported by the Ministry of Science and ICT and the Institute for Information & Communications Technology Planning & Evaluation (IITP) under the “Development of Artificial Intelligence Technology for Personalized Plug-and-Play Explanation and Verification of Explanation” project, the AI Research Hub Project, and the KAIST AI Graduate School Program, and was carried out with support from the Defense Acquisition Program Administration (DAPA) and the Agency for Defense Development (ADD) at the KAIST Center for Applied Research in Artificial Intelligence.
2025.11.26 View 4222 -
Efficient Quantum Process Tomography for Enabling Scalable Optical Quantum Computing
<(From Left) Ph.D candidate Geunhee Gwak, Professor Young-Sik Ra, Dr. Chan Roh, Ph.D candidate Young-Do Yoon from KAIST, (Top Left) Professor M.S Kim from Imperial College London>
Optical quantum computers are gaining attention as a next-generation computing technology with high speed and scalability. However, accurately characterizing complex optical processes, where multiple optical modes interact to generate quantum entanglement, has been considered an extremely challenging task. KAIST research team has overcome this limitation, developing a highly efficient technique that enables complete characterization of complex multimode quantum operations in experiment. This technology, which can analyze large-scale operations with less data, represents an important step toward scalable quantum computing and quantum communication technologies.
KAIST announced on November 17th that a research team led by Professor Young-Sik Ra from the Department of Physics has developed a Multimode Quantum Process Tomography technique capable of efficiently identifying the characteristics of second-order nonlinear optical quantum processes that are essential for optical quantum computing.
Efficient 'CT Scan' Technology for Quantum Computers
'Tomography' is a technique, similar to a medical CT scan, that reconstructs an invisible internal structure from diverse measurements. Similarly, quantum computing requires a method that reconstructs the internal workings of quantum operations using various measurement data. To outperform conventional computers, a quantum computer must be capable of manipulating a large number of quantum units (qubits or qumodes) at the same time. However, as the number of qubits or quantum optical modes (qumodes) increases, the resources required for tomography grows exponentially, making existing technologies unable to analyze systems with even five or more optical modes.
With the newly developed technique, the research team is now able to clearly determine what actually happens inside an optical quantum computer, as if taking a CT scan.
Introducing a New Mathematical Framework Based on Amplification and Noise Matrices
Inside a quantum computer, multiple optical modes interact in a highly complex and entangled way. The research team has introduced a new mathematical framework that precisely describes multimode second-order nonlinear optical quantum processes.
This method analyzes how input states change under a given operation using two key components: the 'Amplification matrix,' which describes how the mean fields of light are transformed, and the 'Noise matrix,' which captures the noise or loss introduced through environmental interactions.
Together, these components create a 'quantum state map' that enables accurate and simultaneous observation of both the ideal quantum evolution of light (unitary changes) and the unavoidable noise (non-unitary changes) present in real devices. This leads to a much more realistic characterization of how an optical quantum computer actually operates.
Reducing the Required Measurement Data and Expanding Analysis to 16 Modes
To determine how a quantum operation works, the research team input several types of quantum states and observed how the outputs changed. They then applied a statistical method known as Maximum Likelihood Estimation to reconstruct the internal operation that most accurately explains the collected data while satisfying the necessary physical conditions.
Using this approach, the research team dramatically reduced the amount of measurement data required. Whereas existing methods quickly become impractical—requiring enormous datasets even for systems with slightly more than a few modes and typically limiting analysis to about five modes—the new technique overcomes this bottleneck. The team successfully performed the world’s first experimental characterization of a large-scale optical quantum operation involving 16 modes, an unprecedented milestone in the field.
<Figure1.Experimental scheme. (Left) Various coherent states are used as input probes to determine the amplification matrix. (Right) A vacuum input state is used to additionally determine the noise matrix.>
<Figure2.Characterization results. (a) 16-mode second-order nonlinear optical quantum process. (b) Cluster state generation. (c) Mode-dependent loss with nonlinear interaction. (d) Quantum noise channel. Left and right columns show the amplification and noise matrices, respectively>
Professor Young-Sik Ra stated, "This research significantly increases the efficiency of Quantum Process Tomography, a foundational technology essential for quantum computing. The acquired technology will greatly contribute to enhancing the scalability and reliability of various quantum technologies, including quantum computing, quantum communication, and quantum sensing."
The study, in which Geunhee Gwak (Integrated M.S, Ph.D. Candidate, Department of Physics) participated as the first author, and Dr. Chan Roh (Postdoctoral Researcher), Young-Do Yoon (Integrated M.S./Ph.D. Candidate), and Professor Myungshik Kim (Imperial College London) participated as co-authors, was formally published online in the prominent international academic journal 'Nature Photonics' on November 11, 2025.
※ Article Title: Completely characterizing multimode second-order nonlinear optical quantum processes, DOI:10.1038/s41566-025-01787-x
This research was supported by the National Research Foundation of Korea (Quantum Computing Technology Development Project, Mid-career Researcher Support Project, Quantum Simulator Development for Material Innovation Project, Quantum Technology R&D Flagship Project, Basic Research Lab Support Project), the Institute of Information & Communications Technology Planning & Evaluation (Core Source Technology for Quantum Internet Project, University ICT Research Center Support Project), and the US Air Force Research Laboratory.
2025.11.18 View 3555 -
Unraveling the Secret of Cell Movement
<(From left) Professor Won Do Heo (KAIST), Postdoctoral Researcher Heeyoung Lee (KAIST, First Author), Professor Kwang-Hyun Cho (KAIST), Professor Kapsang Lee (Johns Hopkins University, USA), Dr. Sangkyu Lee (IBS), Dr. Dongsan Kim (LIBD), Dr. Yeaji Seo (Hulux) (Co-First Authors)>
Cell movement is an essential biological process, whether it's cancer cells metastasizing to other parts of the body or immune cells migrating to heal a wound. However, the principle by which cells autonomously determine their direction of movement without external stimuli has remained unknown until now.
Through this research, KAIST and an international joint research team have elucidated the principle by which cells decide their direction and move on their own, offering a crucial clue for identifying the causes of cancer metastasis and immune diseases and establishing new treatment strategies.
KAIST announced on the 10th of November that the research team led by Endowed Chair Professor Won Do Heo of the Department of Biological Sciences, in collaboration with the research team of Endowed Chair Professor Kwang-Hyun Cho of the Department of Bio and Brain Engineering, and Professor Kapsang Lee's research team at Johns Hopkins University in the US, has for the first time in the world identified the 'autonomous driving mechanism' by which cells determine their direction of movement without external signals.
The research team developed a new imaging technique called 'INSPECT (INtracellular Separation of Protein Engineered Condensation Technique)' that allows direct visualization of how proteins interact within living cells. Using this technology, they revealed the principle of the cell's internal program for autonomously deciding its direction of movement.
The team newly analyzed the operation of the key proteins that regulate cell movement, the Rho family proteins (Rac1, Cdc42, RhoA). The results showed that these proteins do not merely divide the front and back of the cell, as previously theorized, but that the cell's decision to move straight or change direction depends on which protein it binds with.
The INSPECT technology artificially implements the phenomenon of 'phase separation,' where proteins, upon binding, naturally form segregated regions that do not mix well. This technique allows for the direct visualization of how proteins actually bind within the cell using a fluorescent signal.
<Figure 1. INSPECT: A technique for visualizing Intracellular Protein-Protein Interactions">
The research team used the proteins ferritin and the fluorescent protein DsRed to make the clusters, or 'condensates,' visible to the eye when proteins bind together like small droplets.
Using this technology, the team analyzed a total of 285 pairs of interactions by combining 15 types of Rho proteins with 19 types of binding proteins, confirming actual binding in 139 pairs. Specifically, they identified that the Cdc42–FMNL protein combination is the core circuit responsible for the cell's 'straight movement,' while the Rac1–ROCK protein combination is responsible for the cell's 'change of direction.'
The research team slightly modified a part of the Rac1 protein (the 37th amino acid), which is crucial for cell direction control, to prevent it from binding well with the 'steering wheel' protein, ROCK. As a result, the cells could not change direction and continued to move in a straight line.
In contrast, in normal cells, Rac1 and ROCK bind well, forming a structure called 'arc stress fiber' at the front of the cell. This fiber enables the cell to make near-perpendicular turns when changing direction.
Furthermore, in an experiment where the environment the cells were attached to was changed, normal cells adjusted their moving speed according to the surrounding environment, but the Rac1F37W cells (cells with a broken 'steering wheel') maintained the same speed regardless of environmental changes. This demonstrates that the Rac–ROCK protein axis subtly controls the cell's ability to recognize and adapt to its surrounding environment.
<Figure 2. Analysis of the Signaling Network through Screening of Protein Interactions that Bind to a Cell Migration-Controlling Protein>
Professor Won Do Heo stated, "This research reveals that cell movement is not a random motion but is precisely controlled by an intrinsic program created by the ensemble of Rho signaling proteins and cell migration-related proteins." He added, "The newly developed INSPECT technology is a powerful tool for visualizing intracellular protein interactions and will be broadly utilized to uncover the molecular mechanisms of various life phenomena and diseases, such as cancer metastasis and neuronal cell migration."
This research, in which Dr. Heeyoung Lee of KAIST, Dr. Sangkyu Lee (currently at IBS), Dr. Yeji Seo (currently at Hulux Co., Ltd.), and Dr. Dongsan Kim (currently at LIBD) participated as co-first authors, was published in Nature Communications on October 31st.
Journal Name: A Rho GTPase-effector ensemble governs cell migration behavior
DOI: https://doi.org/10.1038/s41467-025-64635-0
The research was supported by the Samsung Future Technology Foundation and the National Research Foundation of Korea.
2025.11.10 View 2528 -
KAIST Discovers Role of Huntingtin Protein in Building the Cell Skeleton
<(From Left) Professor Ji-Joon Song, Ph.D candidate Jaesung Kim, Dr. Hyeongju Kim of KAIST’s Department of Biological Sciences>
Huntington’s disease is a rare genetic disorder and a representative neurodegenerative disease, characterized by loss of motor control, cognitive decline, and psychiatric problems. An international research team has discovered that the “huntingtin protein,” the causal protein of Huntington’s disease (whose mutations are the direct cause of the disease), also performs a new function: directly organizing the cytoskeleton, the fine structural framework inside cells. This discovery is expected to contribute not only to understanding the pathogenic mechanism of Huntington’s disease, but also to research on neurodevelopmental disorders such as Alzheimer’s disease and Parkinson’s disease, as well as muscle- or mobility-related diseases such as muscular dystrophy.
KAIST (President Kwang Hyung Lee) announced on September 30 that a research team led by Professor Ji-Joon Song of the Department of Biological Sciences, in collaboration with the Institute of Science and Technology Austria (ISTA), Sorbonne University/Paris Brain Institute, and the Swiss Federal Institute of Technology Lausanne (EPFL), has uncovered—through cryo-electron microscopy (cryo-EM) and cell biology methods—the structural principle by which the huntingtin protein arranges cytoskeletal microfilaments (F-actin) into bundles.
Until now, the huntingtin protein was known only to “use” the cytoskeleton, being involved in vesicle transport or microtubule-based transport. The team, however, demonstrated that huntingtin physically organizes the cytoskeleton itself. This study is considered the first in the world to prove this new role of the huntingtin protein at the molecular level.
The researchers confirmed that huntingtin binds directly to cytoskeletal microfilaments (F-actin), and that pairs of huntingtin proteins bundle the cytoskeleton into arrays at intervals of about 20 nanometers.
Such cytoskeletal bundles play a crucial role in the development of neural connectivity. Indeed, structural development of neurons was found to be impaired in nerve cells deficient in the huntingtin protein.
<Elucidation of the Mechanism of Cytoskeletal Microfilament Bundle Formation by Huntingtin Protein and Its Impact on Neuronal Development>
First author Jaesung Kim, a PhD candidate at KAIST, stated, “This study provides a new perspective for understanding the molecular mechanism of the huntingtin protein, the cause of an incurable disease that has long remained a mystery.”
Professor Ji-Joon Song of KAIST’s Department of Biological Sciences explained, “This achievement not only provides an important clue to understanding the pathogenic mechanism of Huntington’s disease, but is also expected to have a far-reaching impact on research into cytoskeleton-related diseases,” and added that “it opens new avenues for exploring the role of the huntingtin protein in diverse biological phenomena such as cell division, migration, and mechanical signal transduction.”
This research was conducted with Jaesung Kim (PhD candidate, KAIST), Hyeongju Kim (now at Harvard University), Rémi Carpentier (Paris Brain Institute), Mariacristina Capizzi (Paris Brain Institute), and others as co-first authors, and was published on September 19 in Science Advances, a sister journal of Science.
※ Paper title: “Structure of the Huntingtin F-actin complex reveals its role in cytoskeleton organization,” DOI: https://doi.org/10.1126/sciadv.adw4124※ Co-corresponding authors: Ji-Joon Song (KAIST), Florian Schur (ISTA), and Sandrine Humbert (Sorbonne University/Paris Brain Institute).
This research was supported by the Ministry of Health and Welfare’s Global Research Collaboration Program (Korea–Switzerland Biohealth International Joint Research) and the Korea–Austria Cooperation Program.
2025.09.30 View 3822