Cancer’s secret safety net

Researchers in Class of 1942 Professor of Chemistry Matthew D. Shoulders’ lab have uncovered a sinister hidden mechanism that can allow cancer cells to survive (and, in some cases, thrive) even when hit with powerful drugs. The secret lies in a cellular “safety net” that gives cancer the freedom to develop aggressive mutations.

This fascinating intersection between molecular biology and evolutionary dynamics, published Jan. 22 on the cover of Molecular Cell, focuses on the most famous anti-cancer gene in the human body, TP53 (tumor protein 53, known as p53), and suggests that cancer cells don’t just mutate by accident — they create a specialized environment that makes dangerous mutations possible. 

The guardian under attack

Tasked with the job of stopping damaged cells from dividing, the p53 protein has been known for decades as the “guardian of the genome” and is the most mutated gene in cancer. Some of the most perilous of these mutations are known as “dominant-negative” variants. Not only do they stop working, but they actually prevent any healthy p53 in the cell from doing its job, essentially disarming the body’s primary defense system.

To function, p53 and most other proteins must fold into specific 3D shapes, much like precise cellular origami. Typically, if a mutation occurs that ruins this shape, the protein becomes a tangled mess, and the cell destroys it.

A specialized network of proteins, called cellular chaperones, help proteins fold into their correct shape, collectively known as the proteostasis network. 

“Many chaperone networks are known to be upregulated in cancer cells, for reasons that are not totally clear,” says Stephanie Halim, a graduate student in the Shoulders Group and co-first author of the study, along with Rebecca Sebastian PhD ’22. “We hypothesized that increasing the activities of these helpful protein folding networks can allow cancer cells to tolerate more mutations than a regular cell.”

The research team investigated a “helper” system in the cell called the proteostasis network. This network involves many proteins known as chaperones that help other proteins fold correctly. A master regulator called Heat Shock Factor 1 (HSF1) controls the composition of the proteostasis network, with HSF1 activity upregulating the network to create supportive protein folding environments in response to stress. In healthy cells, HSF1 stays dormant until heat or toxins appear. In cancer, HSF1 is often permanently in action mode.

To see how this works in real-time, the team created a specialized cancer cell line that let them chemically “turn up” the activity of HSF1 on demand. They then used a cutting-edge technique to express every possible singly mutated version of a p53 protein — testing thousands of different genetic “typos” at once.

The results were clear: When HSF1 was amplified, the cancer cells became much better at handling “bad” mutations. Normally, these specific mutations are so physically disruptive that they would cause the protein to collapse and fail. However, with HSF1 providing extra folding help, these unstable, cancer-driving proteins were able to stay intact and keep the cancer growing.

“These findings show that chaperone networks can reshape the fundamental mutational tolerance of the most mutated gene in cancer, linking proteostasis network activity directly to cancer development,” said Halim. “This work also puts us one step closer to understanding how tinkering with cellular protein folding pathways can help with cancer treatment.”

Unravelling cancer’s safety net

The study revealed that HSF1 activity specifically protects normally disruptive amino acid substitutions located deep inside the protein’s core — the most sensitive areas. Without this extra folding help, these substitutions would likely cause degradation of these proteins. With it, the cancer cell can keep these broken proteins around to help it grow.

This discovery helps explain why cancer is so resilient, and why previous attempts to treat cancer by blocking chaperone proteins (like HSP90, an abundant cellular chaperone) have been so complex. By understanding how cancer “buffers” its own bad mutations, doctors may one day be able to break that safety net, forcing the cancer’s own mutations to become its downfall.

The research was conducted in collaboration with the labs of professors Yu-Shan Lin of Tufts University; Francisco J. Sánchez-Rivera of the MIT Department of Biology; William C. Hahn, institute member of the Broad Institute of MIT and Harvard and professor of medicine in the Department of Medical Oncology at the Dana-Farber Cancer Institute and Harvard Medical School; and Marc L. Mendillo of Northwestern University.

de MIT News https://ift.tt/3zlUmf5

Richard Hynes, a pioneer in the biology of cellular adhesion, dies at 81

MIT Professor Emeritus Richard O. Hynes PhD ’71, a cancer biologist whose discoveries reshaped modern understandings of how cells interact with each other and their environment, passed away on Jan. 6. He was 81.

Hynes is best known for his discovery of integrins, a family of cell-surface receptors essential to cell–cell and cell–matrix adhesion. He played a critical role in establishing the field of cell adhesion biology, and his continuing research revealed mechanisms central to embryonic development, tissue integrity, and diseases including cancer, fibrosis, thrombosis, and immune disorders.

Hynes was the Daniel K. Ludwig Professor for Cancer Research, Emeritus, an emeritus professor of biology, and a member of the Koch Institute for Integrated Cancer Research at MIT and the Broad Institute of MIT and Harvard. During his more than 50 years on the faculty at MIT, he was deeply respected for his academic leadership at the Institute and internationally, as well as his intellectual rigor and contributions as an educator and mentor.

“Richard had an enormous impact in his career. He was a visionary leader of the MIT Cancer Center, what is now the Koch Institute, during a time when the progress in understanding cancer was just starting to be translated into new therapies,” reflects Matthew Vander Heiden, director of the Koch Institute and the Lester Wolfe (1919) Professor of Molecular Biology. “The research from his laboratory launched an entirely new field by defining the molecules that mediate interactions between cells and between cells and their environment. This laid the groundwork for better understanding the immune system and metastasis.”

Pond skipper

Born in Kenya, Hynes grew up during the 1950s in Liverpool, in the United Kingdom. While he sometimes recounted stories of being schoolmates with two of the Beatles, and in the same Boy Scouts troop as Paul McCartney, his academic interests were quite different, and he specialized in the sciences at a young age. Both of his parents were scientists: His father was a freshwater ecologist, and his mother a physics teacher. Hynes and all three of his siblings followed their parents into scientific fields.

"We talked science at home, and if we asked questions, we got questions back, not answers. So that conditioned me into being a scientist, for sure," Hynes said of his youth.

He described his time as an undergraduate and master’s student at Cambridge University during the 1960s as “just fantastic,” noting that it was shortly after two 1962 Nobel Prizes were awarded to Cambridge researchers — one to Francis Crick and James Watson for the structure of DNA, the other to John Kendrew and Max Perutz for the structures of proteins — and Cambridge was “the place to be” to study biology.

Newly married, Hynes and his wife traded Cambridge, U.K. for Cambridge, Massachusetts, so that he could conduct doctoral work at MIT under the direction of Paul Gross. He tried (and by his own assessment, failed) to differentiate maternal messages among the three germ layers of sea urchin embryos. However, he did make early successful attempts to isolate the globular protein tubulin, a building block for essential cellular structures, from sea urchins.

Inspired by a course he had taken with Watson in the United States, Hynes began work during his postdoc at the Institute of Cancer Research in the U.K. on the early steps of oncogenic transformation and the role of cell migration and adhesion; it was here that he made his earliest discovery and characterizations of the fibronectin protein.

Recruited back to MIT by Salvador Luria, founding director of the MIT Center for Cancer Research, whom he had met during a summer at Woods Hole Oceanographic Institute on Cape Cod, Hynes returned to the Institute in 1975 as a founding faculty member of the center and an assistant professor in the Department of Biology.

Big questions about tiny cells

To his own research, Hynes brought the same spirit of inquiry that had characterized his upbringing, asking fundamental questions: How do cells interact with each other? How do they stick together to form tissues?

His research focused on proteins that allow cells to adhere to each other and to the extracellular matrix — a mesh-like network that surrounds cells, providing structural support, as well as biochemical and mechanical cues from the local microenvironment. These proteins include integrins, a type of cell surface receptor, and fibronectins, a family of extracellular adhesive proteins. Integrins are the major adhesion receptors connecting the extracellular matrix to the intracellular cytoskeleton, or main architectural support within the cell.

Hynes began his career as a developmental biologist, studying how cells move to the correct locations during embryonic development. During this stage of development, proper modulation of cell adhesion is critical for cells to move to the correct locations in the embryo.

Hynes’ work also revealed that dysregulation of cell-to-matrix contact plays an important role in cancer cells’ ability to detach from a tumor and spread to other parts of the body, key steps in metastasis.

As a postdoc, Hynes had begun studying the differences in the surface landscapes of healthy cells and tumor cells. It was this work that led to the discovery of fibronectin, which is often lost when cells become cancerous.

He and others found that fibronectin is an important part of the extracellular matrix. When fibronectin is lost, cancer cells can more easily free themselves from their original location and metastasize to other sites in the body. By studying how fibronectin normally interacts with cells, Hynes and others discovered a family of cell surface receptors known as integrins, which function as important physical links with the extracellular matrix. In humans, 24 integrin proteins have been identified. These proteins help give tissues their structure, enable blood to clot, and are essential for embryonic development.

“Richard’s discoveries, along with others’, of cell surface integrins led to the development of a number of life-altering treatments. Among these are treatment of autoimmune diseases such as multiple sclerosis,” notes longtime colleague Phillip Sharp, MIT Institute professor emeritus.

As research technologies advanced, including proteomic and extracellular matrix isolation methods developed directly in Hynes’ laboratory, he and his group were able to uncover increasingly detailed information about specific cell adhesion proteins, the biological mechanisms by which they operate, and the roles they play in normal biology and disease.

In cancer, their work helped to uncover how cell adhesion (and the loss thereof) and the extracellular matrix contribute not only to fundamental early steps in the metastatic process, but also tumor progression, therapeutic response, and patient prognosis. This included studies that mapped matrix protein signatures associated with cancer and non-cancer cells and tissues, followed by investigations into how differentially expressed matrix proteins can promote or suppress cancer progression. 

Hynes and his colleagues also demonstrated how extracellular matrix composition can influence immunotherapy, such as the importance of a family of cell adhesion proteins called selectins for recruiting natural killer cells to tumors. Further, Hynes revealed links between fibronectin, integrins, and other matrix proteins with tumor angiogenesis, or blood vessel development, and also showed how interaction with platelets can stimulate tumor cells to remodel the extracellular matrix to support invasion and metastasis. In pursuing these insights into the oncogenic mechanisms of matrix proteins, Hynes and members of his laboratory have identified useful diagnostic and prognostic biomarkers, as well as therapeutic targets.

Along the way, Hynes shaped not only the research field, but also the careers of generations of trainees.

“There was much to emulate in Richard’s gentle, patient, and generous approach to mentorship. He centered the goals and interests of his trainees, fostered an inclusive and intellectually rigorous environment, and cared deeply about the well-being of his lab members. Richard was a role model for integrity in both personal and professional interactions and set high expectations for intellectual excellence,” recalls Noor Jailkhani, a former Hynes Lab postdoc.

Jailkhani is CEO and co-founder, with Hynes, of Matrisome Bio, a biotech company developing first-in-class targeted therapies for cancer and fibrosis by leveraging the extracellular matrix. “The impact of his long and distinguished scientific career was magnified through the generations of trainees he mentored, whose influence spans academia and the biotechnology industry worldwide. I believe that his dedication to mentorship stands among his most far-reaching and enduring contributions,” she says.

A guiding light

Widely sought for his guidance, Hynes served in a number of key roles at MIT and in the broader scientific community. As head of MIT’s Department of Biology from 1989 to 1991, then a decade as director of the MIT Center for Cancer Research, his leadership has helped shape the Institute’s programs in both areas.

“Words can’t capture what a fabulous human being Richard was. I left every interaction with him with new insights and the warm glow that comes from a good conversation,” says Amy Keating, the Jay A. Stein (1968) Professor, professor of biology and biological engineering, and head of the Department of Biology. “Richard was happy to share stories, perspectives, and advice, always with a twinkle in his eye that conveyed his infinite interest in and delight with science, scientists, and life itself. The calm support that he offered me, during my years as department head, meant a lot and helped me do my job with confidence.”

Hynes served as director of the MIT Center for Cancer Research from 1991 until 2001, positioning the center’s distinguished cancer biology program for expansion into its current, interdisciplinary research model as MIT’s Koch Institute for Integrative Cancer Research. “He recruited and strongly supported Tyler Jacks to the faculty, who subsequently became director and headed efforts to establish the Koch Institute,” recalls Sharp.

Jacks, a David H. Koch (1962) Professor of Biology and founding director of the Koch Institute, remembers Hynes as a thoughtful, caring, and highly effective leader in the Center for Cancer Research, or CCR, and in the Department of Biology. “I was fortunate to be able to lean on him when I took over as CCR director. He encouraged me to drop in — unannounced — with questions and concerns, which I did regularly. I learned a great deal from Richard, at every level,” he says.

Hynes’ leadership and recognition extended well beyond MIT to national and international contexts, helping to shape policy and strengthen connections between MIT researchers and the wider field. He served as a scientific governor of the Wellcome Trust, a global health research and advocacy foundation based in the United Kingdom, and co-chaired U.S. National Academy committees establishing guidelines for stem cell and genome editing research.

“Richard was an esteemed scientist, a stimulating colleague, a beloved mentor, a role model, and to me a partner in many endeavors both within and beyond MIT,” notes H. Robert Horvitz, a David H. Koch (1962) Professor of Biology. He was a wonderful human being, and a good friend. I am sad beyond words at his passing.”

Awarded Howard Hughes medical investigator status in 1988, Hynes’ research and leadership have since been recognized with a number of other notable honors. Most recently, he received the 2022 Albert Lasker Basic Medical Research Award, which he shared with Erkki Ruoslahti of Sanford Burnham Prebys and Timothy Springer of Harvard University, for his discovery of integrins and pioneering work in cell adhesion.

His other awards include the Canada Gairdner International Award, the Distinguished Investigator Award from the International Society for Matrix Biology, the Robert and Claire Pasarow Medical Research Award, the E.B. Wilson Medal from the American Society for Cell Biology, the David Rall Medal from the National Academy of Medicine and the Paget-Ewing Award from the Metastasis Research Society. Hynes was a member of the National Academy of Sciences, the National Academy of Medicine, the Royal Society of London, the American Association for the Advancement of Science, and the American Academy of Arts and Sciences.

Easily recognized by a commanding stature that belied his soft-spoken nature, Hynes was known around MIT’s campus not only for his acuity, integrity, and wise counsel, but also for his community spirit and service. From serving food at community socials to moderating events and meetings or recognizing the success of colleagues and trainees, his willingness to help spanned roles of every size.

“Richard was a phenomenal friend and colleague. He approached complex problems with a thoughtfulness and clarity that few can achieve,” notes Vander Heiden. “He was also so generous in his willingness to provide help and advice, and did so with a genuine kindness that was appreciated by everyone.”

Hynes is survived by his wife Fleur, their sons Hugh and Colin and their partners, and four grandchildren.

de MIT News https://ift.tt/1zFfmAg

Featured video: How tiny satellites help us track hurricanes and other weather events

MIT Lincoln Laboratory has transformed weather intelligence by miniaturizing microwave sounders, instruments that measure Earth's atmospheric temperature, moisture, and water vapor. These instruments are 1/100th the size of traditional sounders aboard multibillion-dollar satellites, enabling them to fit on shoebox-sized CubeSats. 

When deployed in a constellation, the CubeSats can observe rapidly intensifying storms near-hourly — providing fresh data to forecasting professionals during critical windows of storm development that have largely been undetectable by past remote-sensing technology.

Developed at Lincoln Laboratory, the mini microwave sounders were first demonstrated on NASA's TROPICS mission, which measured temperature and humidity soundings as well as precipitation. TROPICS concluded in 2025 with over 11 billion observations, providing scientists with key insights into tropical cyclone evolution. 

Now the technology has been licensed by the commercial firm Tomorrow.io, allowing for the enhancement of global weather coverage for customers in aviation, logistics, agriculture, and emergency management. Tomorrow.io provides clients with hyperlocal forecasts around the globe and is set to launch their own constellation of satellites based on the TROPICS program. Says John Springman, Tomorrow.io's head of space and sensing: “Our overall goal is to fundamentally improve weather forecasts, and that'll improve our downstream products like our weather intelligence.”

Video by Tim Briggs/Lincoln Laboratory | 13 minutes, 58 seconds

de MIT News https://ift.tt/DyMLed3

Professor of the practice Robert Liebeck, leading expert on aircraft design, dies at 87

Robert Liebeck, a professor of the practice in the MIT Department of Aeronautics and Astronautics and one of the world’s leading experts on aircraft design, aerodynamics, and hydrodynamics, died on Jan. 12 at age 87.

“Bob was a mentor and dear friend to so many faculty, alumni, and researchers at AeroAstro over the course of 25 years,” says Julie Shah, department head and the H.N. Slater Professor of Aeronautics and Astronautics at MIT. “He’ll be deeply missed by all who were fortunate enough to know him.”

Liebeck’s long and distinguished career in aerospace engineering included a number of foundational contributions to aerodynamics and aircraft design, beginning with his graduate research into high-lift airfoils. His novel designs came to be known as “Liebeck airfoils” and are used primarily for high-altitude reconnaissance airplanes; Liebeck airfoils have also been adapted for use in Formula One racing cars, racing sailboats, and even a flying replica of a giant pterosaur.

He was perhaps best known for his groundbreaking work on blended wing body (BWB) aircraft. He oversaw the BWB project at Boeing during his celebrated five-decade tenure at the company, working closely with NASA on the X-48 experimental aircraft. After retiring as senior technical fellow at Boeing in 2020, Liebeck remained active in BWB research. He served as technical advisor at BWB startup JetZero, which is aiming to build a more fuel-efficient aircraft for both military and commercial use and has set a target date of 2027 for its demonstration flight. 

Liebeck was appointed a professor of the practice at MIT in 2000, and taught classes on aircraft design and aerodynamics. 

“Bob contributed to the department both in aircraft capstones and also in advising students and mentoring faculty, including myself,” says John Hansman, the T. Wilson Professor of Aeronautics and Astronautics. “In addition to aviation, Bob was very significant in car racing and developed the downforce wing and flap system which has become standard on F1, IndyCar, and NASCAR cars.”

He was a major contributor to the Silent Aircraft Project, a collaboration between MIT and Cambridge University led by Dame Ann Dowling. Liebeck also worked closely with Professor Woody Hoburg ’08 and his research group, advising on students’ research into efficient methods for designing aerospace vehicles. Before Hoburg was accepted into the NASA astronaut corps in 2017, the group produced an open-source Python package, GPkit, for geometric programming, which was used to design a five-day endurance unmanned aerial vehicle for the U.S. Air Force.

“Bob was universally respected in aviation and he was a good friend to the department,” remembers Professor Ed Greitzer.

Liebeck was an AIAA honorary fellow and Boeing senior technical fellow, as well as a member of the National Academy of Engineering, Royal Aeronautical Society, and Academy of Model Aeronautics. He was a recipient of the Guggenheim Medal and ASME Spirit of St. Louis Medal, among many other awards, and was inducted into the International Air and Space Hall of Fame.

An avid runner and motorcyclist, Liebeck is remembered by friends and colleagues for his adventurous nature and generosity of spirit. Throughout a career punctuated by honors and achievements, Liebeck found his greatest satisfaction in teaching. In addition to his role at MIT, he was an adjunct faculty member at University of California at Irving and served as faculty member for that university’s Design/Build/Fly and Human-Powered Airplane teams.

“It is the one job where I feel I have done some good — even after a bad lecture,” he told AeroAstro Magazine in 2007. “I have decided that I am finally beginning to understand aeronautical engineering, and I want to share that understanding with our youth.”

de MIT News https://ift.tt/We1rNnv

Electrifying boilers to decarbonize industry

More than 200 years ago, the steam boiler helped spark the Industrial Revolution. Since then, steam has been the lifeblood of industrial activity around the world. Today the production of steam — created by burning gas, oil, or coal to boil water — accounts for a significant percentage of global energy use in manufacturing, powering the creation of paper, chemicals, pharmaceuticals, food, and more.

Now, the startup AtmosZero, founded by Addison Stark SM ’10, PhD ’14; Todd Bandhauer; and Ashwin Salvi, is taking a new approach to electrify the centuries-old steam boiler. The company has developed a modular heat pump capable of delivering industrial steam at temperatures up to 150 degrees Celsius to serve as a drop-in replacement for combustion boilers.

The company says its first 1-megawatt steam system is far cheaper to operate than commercially available electric solutions thanks to ultra-efficient compressor technology, which uses 50 percent less electricity than electric resistive boilers. The founders are hoping that’s enough to make decarbonized steam boilers drive the next industrial revolution.

“Steam is the most important working fluid ever,” says Stark, who serves as AtmosZero’s CEO. “Today everything is built around the ubiquitous availability of steam. Cost-effectively electrifying that requires innovation that can scale. In other words, it requires a mass-produced product — not one-off projects.”

Tapping into steam

Stark joined the Technology and Policy Program when he came to MIT in 2007. He ultimately completed a dual master’s degree by adding mechanical engineering to his studies.

“I was interested in the energy transition and in accelerating solutions to enable that,” Stark says. “The transition isn’t happening in a vacuum. You need to align economics, policy, and technology to drive that change.”

Stark stayed at MIT to earn his PhD in mechanical engineering, studying thermochemical biofuels.

After MIT, Stark began working on early-stage energy technologies with the Department of Energy’s Advanced Research Projects Agency— Energy (ARPA-E), with a focus on manufacturing efficiency, the energy-water nexus, and electrification.

“Part of that work involved applying my training at MIT to things that hadn’t really been innovated on for 50 years,” Stark says. “I was looking at the heat exchanger. It’s so fundamental. I thought, ‘How might we reimagine it in the context of modern advances in manufacturing technology?’”

The problem is as difficult as it is consequential, touching nearly every corner of the global industrial economy. More than 2.2 gigatons of CO₂ emissions are generated each year to turn water into steam — accounting for more than 5 percent of global energy-related emissions.

In 2020, Stark co-authored an article in the journal Joule with Gregory Thiel SM ’12, PhD ’15 titled, “To decarbonize industry, we must decarbonize heat.” The article examined opportunities for industrial heat decarbonization, and it got Stark excited about the potential impact of a standardized, scalable electric heat pump.

Most electric boiler options today bring huge increases in operating costs. Many also make use of factory waste heat, which requires pricey retrofits. Stark never imagined he’d become an entrepreneur, but he soon realized no one was going to act on his findings for him.

“The only path to seeing this invention brought out into the world was to found and run the company,” Stark says. “It’s something I didn’t anticipate or necessarily want, but here I am.”

Stark partnered with former ARPA-E awardee Todd Bandhauer, who had been inventing new refrigerant compressor technology in his lab at Colorado State University, and former ARPA-E colleague Ashwin Salvi. The team officially founded AtmosZero in 2022.

“The compressor is the engine of the heat pump and defines the efficiency, cost, and performance,” Stark says. “The fundamental challenge of delivering heat is that the higher your heat pump is raising the air temperature, the lower your maximum efficiency. It runs into thermodynamic limitations. By designing for optimum efficiency in the operational windows that matter for the refrigerants we’re using, and for the precision manufacturing of our compressors, we’re able to maximize the individual stages of compression to maximize operational efficiency.”

The system can work with waste heat from air or water, but it doesn’t need waste heat to work. Many other electric boilers rely on waste heat, but Stark thinks that adds too much complexity to installation and operations.

Instead, in AtmosZero’s novel heat pump cycle, heat from ambient-temperature air is used to warm a liquid heat transfer material, which evaporates a refrigerant so it flows into the system’s series of compressors and heat exchangers, reaching high enough temperatures to boil water while recovering heat from the refrigerant once it reaches lower temperatures. The system can be ramped up and down to seamlessly fit into existing industrial processes.

“We can work just like a combustion boiler,” Stark says. “At the end of the day, customers don’t want to change how their manufacturing facilities operate in order to electrify. You can’t change or increase complexity on-site.”

That approach means the boiler can be deployed in a range of industrial contexts without unique project costs or other changes.

“What we really offer is flexibility and something that can drop in with ease and minimize total capital costs,” Stark says.

From 1 to 1,000

AtmosZero already has a pilot 650 kilowatt system operating at a customer facility near its headquarters in Loveland, Colorado. The company is currently focused on demonstrating year-round durability and reliability of the system as they work to build out their backlog of orders and prepare to scale. 

Stark says once the system is brought to a customer’s facility, it can be installed in an afternoon and deployed in a matter of days, with zero downtime.

AtmosZero is aiming to deliver a handful of units to customers over the next year or two, with plans to deploy hundreds of units a year after that. The company is currently targeting manufacturing plants using under 10 megawatts of thermal energy at peak demand, which represents most U.S. manufacturing facilities.

Stark is proud to be part of a growing group of MIT-affiliated decarbonization startups, some of which are targeting specific verticals, like Boston Metal for steel and Sublime Systems for cement. But he says beyond the most common materials, the industry gets very fragmented, with one of the only common threads being the use of steam.

“If we look across industrial segments, we see the ubiquity of steam,” Stark says. “It’s a tremendously ripe opportunity to have impact at scale. Steam cannot be removed from industry. So much of every industrial process that we’ve designed over the last 160 years has been around the availability of steam. So, we need to focus on ways to deliver low-emissions steam rather than removing it from the equation.”

de MIT News https://ift.tt/t5zyHen

Why it’s critical to move beyond overly aggregated machine-learning metrics

MIT researchers have identified significant examples of machine-learning model failure when those models are applied to data other than what they were trained on, raising questions about the need to test whenever a model is deployed in a new setting.

“We demonstrate that even when you train models on large amounts of data, and choose the best average model, in a new setting this ‘best model’ could be the worst model for 6-75 percent of the new data,” says Marzyeh Ghassemi, an associate professor in MIT’s Department of Electrical Engineering and Computer Science (EECS), a member of the Institute for Medical Engineering and Science, and principal investigator at the Laboratory for Information and Decision Systems.

In a paper that was presented at the Neural Information Processing Systems (NeurIPS 2025) conference in December, the researchers point out that models trained to effectively diagnose illness in chest X-rays at one hospital, for example, may be considered effective in a different hospital, on average. The researchers’ performance assessment, however, revealed that some of the best-performing models at the first hospital were the worst-performing on up to 75 percent of patients at the second hospital, even though when all patients are aggregated in the second hospital, high average performance hides this failure.

Their findings demonstrate that although spurious correlations — a simple example of which is when a machine-learning system, not having “seen” many cows pictured at the beach, classifies a photo of a beach-going cow as an orca simply because of its background — are thought to be mitigated by just improving model performance on observed data, they actually still occur and remain a risk to a model’s trustworthiness in new settings. In many instances — including areas examined by the researchers such as chest X-rays, cancer histopathology images, and hate speech detection — such spurious correlations are much harder to detect.

In the case of a medical diagnosis model trained on chest X-rays, for example, the model may have learned to correlate a specific and irrelevant marking on one hospital’s X-rays with a certain pathology. At another hospital where the marking is not used, that pathology could be missed.

Previous research by Ghassemi’s group has shown that models can spuriously correlate such factors as age, gender, and race with medical findings. If, for instance, a model has been trained on more older people’s chest X-rays that have pneumonia and hasn’t “seen” as many X-rays belonging to younger people, it might predict that only older patients have pneumonia.

“We want models to learn how to look at the anatomical features of the patient and then make a decision based on that,” says Olawale Salaudeen, an MIT postdoc and the lead author of the paper, “but really anything that’s in the data that’s correlated with a decision can be used by the model. And those correlations might not actually be robust with changes in the environment, making the model predictions unreliable sources of decision-making.”

Spurious correlations contribute to the risks of biased decision-making. In the NeurIPS conference paper, the researchers showed that, for example, chest X-ray models that improved overall diagnosis performance actually performed worse on patients with pleural conditions or enlarged cardiomediastinum, meaning enlargement of the heart or central chest cavity.

Other authors of the paper included PhD students Haoran Zhang and Kumail Alhamoud, EECS Assistant Professor Sara Beery, and Ghassemi.

While previous work has generally accepted that models ordered best-to-worst by performance will preserve that order when applied in new settings, called accuracy-on-the-line, the researchers were able to demonstrate examples of when the best-performing models in one setting were the worst-performing in another.

Salaudeen devised an algorithm called OODSelect to find examples where accuracy-on-the-line was broken. Basically, he trained thousands of models using in-distribution data, meaning the data were from the first setting, and calculated their accuracy. Then he applied the models to the data from the second setting. When those with the highest accuracy on the first-setting data were wrong when applied to a large percentage of examples in the second setting, this identified the problem subsets, or sub-populations. Salaudeen also emphasizes the dangers of aggregate statistics for evaluation, which can obscure more granular and consequential information about model performance.

In the course of their work, the researchers separated out the “most miscalculated examples” so as not to conflate spurious correlations within a dataset with situations that are simply difficult to classify.

The NeurIPS paper releases the researchers’ code and some identified subsets for future work.

Once a hospital, or any organization employing machine learning, identifies subsets on which a model is performing poorly, that information can be used to improve the model for its particular task and setting. The researchers recommend that future work adopt OODSelect in order to highlight targets for evaluation and design approaches to improving performance more consistently.

“We hope the released code and OODSelect subsets become a steppingstone,” the researchers write, “toward benchmarks and models that confront the adverse effects of spurious correlations.”

de MIT News https://ift.tt/fm7iFCu

A new way to “paint with light” to create radiant, color-changing items

Gemstones like precious opal are beautiful to look at and deceivingly complex. As you look at such gems from different angles, you’ll see a variety of tints glisten, causing you to question what color the rock actually is. It’s iridescent thanks to something called structural color — microscopic structures that reflect light to produce radiant hues.

Structural color can be found across different organisms in nature, such as on the tails of peacocks and the wings of certain butterflies. Scientists and artists have been working to replicate this quality, but outside of the lab, it’s still very hard to recreate, causing a barrier to on-demand, customizable fabrication. Instead, companies and individual designers alike have resorted to adding existing color-changing objects like feathers and gems to things like personal items, clothes, and artwork.

Now MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) researchers have replicated nature’s brilliance with a new optical system called “MorphoChrome.” MorphoChrome allows users to design and program iridescence onto everyday objects (like a glove, for example), augmenting them with the structurally colored multi-color glimmer reminiscent of many gemstones. You select particular colors from a color wheel in the team’s software program and use their handheld device to “paint” with multi-color light onto holographic film. Then, you apply that painted sheet to 3D-printed objects or flexible substrates such as fashion items, sporting goods, and other personal accessories, using their unique epoxy resin transfer process.

“We wanted to tap into the innate intelligence of nature,” says MIT Department of Electrical Engineering and Computer Science (EECS) PhD student and CSAIL researcher Paris Myers SM ’25, who is a lead author on a recent paper presenting MorphoChrome. “In the past, you couldn’t easily synthesize structural color yourself, but using pigments or dyes gave you full creative expression. With our system, you have full creative agency over this new material space, predictably programming iridescent designs in real-time.”

MorphoChrome showed it could add a luminous touch to things like a necklace charm of a butterfly. What started as a static, black accessory became a shiny pendant with green, orange, and blue glimmers, thanks to the system’s programmable color process. MorphoChrome also turned golfing gloves into beginner-friendly training equipment that shine green when you hold a golf club at the correct angle, and even helped one user adorn their fingernails with a gemstone-like look.

These multi-color displays are the result of a handheld fabrication process where MorphoChrome acts as a “brush" to paint with red-green-blue (RGB) laser light, while a holographic photopolymer film (used for things like passports and debit cards) is the canvas. Users first connect the system’s handheld device to a computer via a USB-C port, then open the software program. They can then click “send color” to rapidly transmit different hues from their laptop or home computer to the MorphoChrome hardware tool.

This handheld device transforms the colors on a screen into a controllable, multi-color RGB laser light output that instantly exposes the film, a sort of canvas where users can explore different combinations of hues. About the size of a glue bottle, MorphoChrome’s optical machine houses red, green, and blue lasers, which are activated at various intensities depending on the color chosen. These lights are reflected off mirrors toward an optical prism, where the colors mix and are promptly released as a single combined beam of light. 

After designing the film, one can fabricate diverse structurally colored objects by first coating a chosen object with a thin layer of epoxy resin. Next, the holographic film (litiholographics) — composed of a photopolymer layer and a protective plastic backing — is bonded to the object through a 20-second ultraviolet cure, essentially using a handheld UV light to transfer the colored design onto the surface. Finally, users peel off the film’s protective plastic sheet, revealing a color-changing, structurally-colored object that looks like a jewel. 

Do try this at home

MorphoChrome is surprisingly user-friendly, consisting of a straightforward fabrication blueprint and an easy-to-use device that encourages do-it-yourself designers and other makers to explore iridescent designs at home. Instead of spending time searching for hard-to-find artistic materials or chemically synthesizing structural color in the lab, users can focus on expressing various ideas and experimenting with programming different radiant color mixes.

The array of possible colors stems from intriguing fusions. Nagenta, for instance, is created after the system’s blue and red lasers mix. Selecting cyan on the MorphoChrome software’s color wheel will mix the green and blue lights.

Users should note that the time it takes to fully expose the film to each color will vary, based on the researchers’ multi-color findings and the intrinsic properties of holographic photopolymer film. MorphoChrome activates green in 2.5 seconds, whereas red takes about 3 seconds, and blue needs roughly 6 seconds to saturate. The reason for this discrepancy is that each color is a particular wavelength of light, requiring a certain level of light exposure (blue needing more than green or red).

Look at this hologram

MorphoChrome builds upon previous work on stretchable structural color by co-author Benjamin Miller PhD ’24, Professor Mathias Kolle, and Kolle’s Laboratory for Biologically Inspired Photonic Engineering group at MIT's Department of Mechanical Engineering. The CSAIL researchers, who work in the Human-Computer Interaction Engineering Group, say that MorphoChrome also advances their ongoing work on merging computation with unique materials to create dynamic, programmable color interfaces. 

Going forward, their goal is to push the capabilities of holographic structural color as a reprogrammable design and manufacturing space, empowering individuals and industries alike to customize iridescent and diffuse multi-color interfaces. “The polymer sheet we went with here is holographic, which has potential beyond what we’re showing here,” says co-author Yunyi Zhu ’20, MEng ’21, who is an MIT EECS PhD student and CSAIL researcher. “We’re working on adapting our process for creating entire 3D light fields in one film.”

Customizing full light-field holographic messages onto objects would allow users to encode information and 3D images. One could imagine, for example, that a passport could have a sticker that beams out a 3D green check mark. This hologram would signal its authenticity when viewed through a particular device or at a certain angle.

The team is also inspired by how animals use structural color as an adaptive communication channel and camouflage technique. Going forward, they are curious how programmable structural color could be integrated into different types of environments, perhaps as camouflage for soft robotic structures to blend into an environment. For instance, they imagine a robot studying jungle terrain may need to match the appearance of nearby bushes to collect data, with a human reprogramming the machine’s color from afar.

In the meantime, MorphoChrome recreates the majestic structural color found in various ecosystems, connecting a natural phenomenon with our creative processes. MIT researchers will look to improve the system’s color gamut and maximize how luminous mixed colors are. They’re also considering using another material for the device’s casing, since its current 3D-printing housing leaks out some light.

“Being able to easily create and manipulate structural color is a great new tool, and opens up new avenues for discovery and expression,” says Liti Holographics CEO Paul Christie SM ’97, who wasn’t involved in the research. “Simplifying the process to be more easily accessible allows for new applications to be developed in a wider range of areas, from art and jewelry to functional fabric.”

Myers, Zhu, and Miller wrote the paper with senior author Stefanie Mueller, who is an MIT associate professor of electrical engineering and computer science and CSAIL principal investigator. Their research was supported by the National Science Foundation, and presented as a demo paper and poster at the 2025 ACM Symposium on Computational Fabrication in November.

de MIT News https://ift.tt/ruRCOAf