Superconductors are like the express trains in a metro system. Any electricity that “boards” a superconducting material can zip through it without stopping and losing energy along the way. As such, superconductors are extremely energy efficient, and are used today to power a variety of applications, from MRI machines to particle accelerators.

But these “conventional” superconductors are somewhat limited in terms of uses because they must be brought down to ultra-low temperatures using elaborate cooling systems to keep them in their superconducting state. If superconductors could work at higher, room-like temperatures, they would enable a new world of technologies, from zero-energy-loss power cables and electricity grids to practical quantum computing systems. And so scientists at MIT and elsewhere are studying “unconventional” superconductors — materials that exhibit superconductivity in ways that are different from, and potentially more promising than, today’s superconductors.

In a promising breakthrough, MIT physicists have today reported their observation of new key evidence of unconventional superconductivity in “magic-angle” twisted tri-layer graphene (MATTG) — a material that is made by stacking three atomically-thin sheets of graphene at a specific angle, or twist, that then allows exotic properties to emerge.

MATTG has shown indirect hints of unconventional superconductivity and other strange electronic behavior in the past. The new discovery, reported in the journal Science, offers the most direct confirmation yet that the material exhibits unconventional superconductivity.

In particular, the team was able to measure MATTG’s superconducting gap — a property that describes how resilient a material’s superconducting state is at given temperatures. They found that MATTG’s superconducting gap looks very different from that of the typical superconductor, meaning that the mechanism by which the material becomes superconductive must also be different, and unconventional.

“There are many different mechanisms that can lead to superconductivity in materials,” says study co-lead author Shuwen Sun, a graduate student in MIT’s Department of Physics. “The superconducting gap gives us a clue to what kind of mechanism can lead to things like room-temperature superconductors that will eventually benefit human society.”

The researchers made their discovery using a new experimental platform that allows them to essentially “watch” the superconducting gap, as the superconductivity emerges in two-dimensional materials, in real-time. They plan to apply the platform to further probe MATTG, and to map the superconducting gap in other 2D materials — an effort that could reveal promising candidates for future technologies.

“Understanding one unconventional superconductor very well may trigger our understanding of the rest,” says Pablo Jarillo-Herrero, the Cecil and Ida Green Professor of Physics at MIT and the senior author of the study. “This understanding may guide the design of superconductors that work at room temperature, for example, which is sort of the Holy Grail of the entire field.”

The study’s other co-lead author is Jeong Min Park PhD ’24; Kenji Watanabe and Takashi Taniguchi of the National Institute for Materials Science in Japan are also co-authors.

The ties that bind

Graphene is a material that comprises a single layer of carbon atoms that are linked in a hexagonal pattern resembling chicken wire. A sheet of graphene can be isolated by carefully exfoliating an atom-thin flake from a block of graphite (the same stuff of pencil lead). In the 2010s, theorists predicted that if two graphene layers were stacked at a very special angle, the resulting structure should be capable of exotic electronic behavior.

In 2018, Jarillo-Herrero and his colleagues became the first to produce magic-angle graphene in experiments, and to observe some of its extraordinary properties. That discovery sprouted an entire new field known as “twistronics,” and the study of atomically thin, precisely twisted materials. Jarillo-Herrero’s group has since studied other configurations of magic-angle graphene with two, three, and more layers, as well as stacked and twisted structures of other two-dimensional materials. Their work, along with other groups, have revealed some signatures of unconventional superconductivity in some structures.

Superconductivity is a state that a material can exhibit under certain conditions (usually at very low temperatures). When a material is a superconductor, any electrons that pass through can pair up, rather than repelling and scattering away. When they couple up in what is known as “Cooper pairs,” the electrons can glide through a material without friction, instead of knocking against each other and flying away as lost energy. This pairing up of electrons is what enables superconductivity, though the way in which they are bound can vary.

“In conventional superconductors, the electrons in these pairs are very far away from each other, and weakly bound,” says Park. “But in magic-angle graphene, we could already see signatures that these pairs are very tightly bound, almost like a molecule. There were hints that there is something very different about this material.”

Tunneling through

In their new study, Jarillo-Herrero and his colleagues aimed to directly observe and confirm unconventional superconductivity in a magic-angle graphene structure. To do so, they would have to measure the material’s superconducting gap.

“When a material becomes superconducting, electrons move together as pairs rather than individually, and there’s an energy ‘gap’ that reflects how they’re bound,” Park explains. “The shape and symmetry of that gap tells us the underlying nature of the superconductivity.”

Scientists have measured the superconducting gap in materials using specialized techniques, such as tunneling spectroscopy. The technique takes advantage of a quantum mechanical property known as “tunneling.” At the quantum scale, an electron behaves not just as a particle, but also as a wave; as such, its wave-like properties enable an electron to travel, or “tunnel,” through a material, as if it could move through walls.

Such tunneling spectroscopy measurements can give an idea of how easy it is for an electron to tunnel into a material, and in some sense, how tightly packed and bound the electrons in the material are. When performed in a superconducting state, it can reflect the properties of the superconducting gap. However, tunneling spectroscopy alone cannot always tell whether the material is, in fact, in a superconducting state. Directly linking a tunneling signal to a genuine superconducting gap is both essential and experimentally challenging.

In their new work, Park and her colleagues developed an experimental platform that combines electron tunneling with electrical transport — a technique that is used to gauge a material’s superconductivity, by sending current through and continuously measuring its electrical resistance (zero resistance signals that a material is in a superconducting state).

The team applied the new platform to measure the superconducting gap in MATTG. By combining tunneling and transport measurements in the same device, they could unambiguously identify the superconducting tunneling gap, one that appeared only when the material exhibited zero electrical resistance, which is the hallmark of superconductivity. They then tracked how this gap evolved under varying temperature and magnetic fields. Remarkably, the gap displayed a distinct V-shaped profile, which was clearly different from the flat and uniform shape of conventional superconductors.

This V shape reflects a certain unconventional mechanism by which electrons in MATTG pair up to superconduct. Exactly what that mechanism is remains unknown. But the fact that the shape of the superconducting gap in MATTG stands out from that of the typical superconductor provides key evidence that the material is an unconventional superconductor.

In conventional superconductors, electrons pair up through vibrations of the surrounding atomic lattice, which effectively jostle the particles together. But Park suspects that a different mechanism could be at work in MATTG.

“In this magic-angle graphene system, there are theories explaining that the pairing likely arises from strong electronic interactions rather than lattice vibrations,” she posits. “That means electrons themselves help each other pair up, forming a superconducting state with special symmetry.”

Going forward, the team will test other two-dimensional twisted structures and materials using the new experimental platform.

“This allows us to both identify and study the underlying electronic structures of superconductivity and other quantum phases as they happen, within the same sample,” Park says. “This direct view can reveal how electrons pair and compete with other states, paving the way to design and control new superconductors and quantum materials that could one day power more efficient technologies or quantum computers.”

This research was supported, in part, by the U.S. Army Research Office, the U.S. Air Force Office of Scientific Research, the MIT/MTL Samsung Semiconductor Research Fund, the Sagol WIS-MIT Bridge Program, the National Science Foundation, the Gordon and Betty Moore Foundation, and the Ramon Areces Foundation.

EFF has, for many years, raised the alarm about the proliferation of stalkerware—commercially-available apps designed to be installed covertly on another person’s device and exfiltrate data from that device without their knowledge. In particular, we have urged the makers of anti-virus products for Android phones to improve their detection of stalkerware and call it out explicitly to users when it is found. In 2020  and 2021, AV Comparatives ran tests to see how well the most popular anti-virus products detected stalkerware from many different vendors. The results were mixed, with some high-scoring companies and others that had alarmingly low detection rates. Since malware detection is an endless game of cat and mouse between anti-virus companies and malware developers, we felt that the time was right to take a more up-to-date snapshot of how well the anti-virus companies are performing. We’ve teamed up with the researchers at AV Comparatives to test the most popular anti-virus products for Android to see how well they detect the most popular stalkerware products in 2025.

Here is what we found:

Stalkerware detection is still a mixed bag. Notably, Malwarebytes detected 100% of the stalkerware products we tested for. ESET, Bitdefender, McAfee, and Kaspersky detected all but one sample. This is a marked improvement over the 2021 test, which also found only one app with a 100% detection rate (G Data), but the next-best performing products had detect rates of 80-85%. Google Play Protect and Trend Micro had the lowest detection rates in the 2025 test, at 53% and 59% respectively. The poor performance of Google Play Protect is unsurprising: because it is the anti-virus solution on so many Android phones by default, some stalkerware includes specific instructions to disable detection by Google Play Protect as part of the installation process.

There are fewer stalkerware products out there. In 2020 and 2021, AV Comparatives tested 20 unique stalkerware products from different vendors. In 2025, we tested 17. We found that many stalkerware apps are essentially variations on the same underlying product and that the number of unique underlying products appears to have decreased in recent years. We cannot be certain about the cause of this decline, but we speculate that increased attention from regulators may be a factor. The popularity of small, cheap, Bluetooth-enabled physical trackers such as Apple AirTags and Tiles as an alternative method of location-tracking may also be undercutting the stalkerware market. 

We hope that these tests will help survivors of domestic abuse and others who are concerned about stalkerware on their Android devices make informed choices about their anti-virus apps. We also hope that exposing the gaps that these products have in stalkerware detection will renew interest in this problem at anti-virus companies.

You can find the full results of the test here (PDF). 

The Department of Justice has indicted thirty-one people over the high-tech rigging of high-stakes poker games.

In a typical legitimate poker game, a dealer uses a shuffling machine to shuffle the cards randomly before dealing them to all the players in a particular order. As set forth in the indictment, the rigged games used altered shuffling machines that contained hidden technology allowing the machines to read all the cards in the deck. Because the cards were always dealt in a particular order to the players at the table, the machines could determine which player would have the winning hand. This information was transmitted to an off-site member of the conspiracy, who then transmitted that information via cellphone back to a member of the conspiracy who was playing at the table, referred to as the “Quarterback” or “Driver.” The Quarterback then secretly signaled this information (usually by prearranged signals like touching certain chips or other items on the table) to other co-conspirators playing at the table, who were also participants in the scheme. Collectively, the Quarterback and other players in on the scheme (i.e., the cheating team) used this information to win poker games against unwitting victims, who sometimes lost tens or hundreds of thousands of dollars at a time. The defendants used other cheating technology as well, such as a chip tray analyzer (essentially, a poker chip tray that also secretly read all cards using hidden cameras), an x-ray table that could read cards face down on the table, and special contact lenses or eyeglasses that could read pre-marked cards. ...

The president's megalaw is making it difficult for cash-strapped hospitals to gird for storms that are being turbocharged by climate change.

Pulling out of the Paris climate agreement was not enough, an ex-aide to the U.S. president says on the eve of a summit in Brazil: "You have to potentially destroy it."

Successful candidates in Virginia, New Jersey and Georgia focused on high electricity bills at a time when utilities are struggling to keep up with data center demand.

A cross section of manufacturers and environmentalists raised objections to the agency's plan to repeal its greenhouse gas reporting rule.

The agency had suspended nearly all activity due to the government shutdown.

It's "unfortunate that this is the take-home message," one scientist said.

The deal avoids the EU ministers going empty-handed to the climate summit that starts Nov. 10.

Britain is “incredibly supportive” of the initiative pushed by COP’s Brazilian hosts — but won’t be stumping up public money.

Europe’s far-right parties are capitalizing on the public’s reluctance to change technologies.

The head of a carbon removal startup explains why Trump climate policy has created a “golden opportunity” for Canada.

Traditional folk ballads are one of our most enduring forms of cultural expression. They can also be lost to society, forgotten over time. That’s why, in the mid-1700s, when a Scottish woman named Anna Gordon was found to know three dozen ancient ballads, collectors tried to document all of these songs — a volume of work that became a kind of sensation in its time, a celebrated piece of cultural heritage.

That story is told in MIT Professor Emerita Ruth Perry’s latest book, “The Ballad World of Anna Gordon, Mrs. Brown of Falkland,” published this year by Oxford University Press. In it, Perry details what we know about the ways folk ballads were created and transmitted; how Anna Gordon came to know so many; the social and political climate in which they existed; and why these songs meant so much in Scotland and elsewhere in the Atlantic world. Indeed, Scottish immigrants brought their music to the U.S., among other places.

MIT News sat down with Perry, who is MIT’s Ann Fetter Friedlaender Professor of Humanities, Emerita, to talk about the book.

Q: This is fascinating topic with a lot of threads woven together. To you, what is the book about?

A: It’s really three books. It’s a book about Anna Gordon and her family, a very interesting middle-class family living in Aberdeen in the middle of the 18th century. And it’s a book about balladry and what a ballad is — a story told in song, and ballads are the oldest known poetry in English. Some of them are gorgeous. Third, it’s a book about the relationship between Scotland and England, the effects of the Jacobite uprising in 1745, social attitudes, how people lived, what they ate, education — it’s very much about 18th century Scotland.

Q: Okay, who was Anna Gordon, and what was her family milieu?

A: Anna’s father, Thomas Gordon, was a professor at King’s College, now the University of Aberdeen. He was a professor of humanity, which in those days meant Greek and Latin, and was well-connected to the intellectual community of the Scottish Enlightenment. A friend of his, an Edinburgh writer, lawyer, and judge, William Tytler, who heard cases all over the country and always stayed with Thomas Gordon and his family when he came to Aberdeen, was intensely interested in Scottish traditional music. He found out that Anna Gordon had learned all these ballads as a child, from her mother and aunt and some servants. Tytler asked if she would write them down, both tunes and words.

That was the earliest manuscript of ballads ever collected from a named person in Scotland. Once it was in existence, all kinds of people wanted to see it; it got spread throughout the country. In my book, I detail much of the excitement over this manuscript.

The thing about Anna’s ballads is: It’s not just that there are more of them, and more complete versions that are fuller, with more verses. They’re more beautiful. The language is more archaic, and there are marvelous touches. It is thought, and I agree, that Anna Gordon was an oral poet. As she remembered ballads and reproduced them, she improved on them. She had a great memory for the best bits and would improve other parts.

Q: How did it come about that at this time, a woman such as Anna Gordon would be the keeper and creator of cultural knowledge?

A: Women were more literate in Scotland than elsewhere. The Scottish Parliament passed an act in 1695 requiring every parish in the Church of Scotland to have not only a minister, but a teacher. Scotland was the most literate country in Europe in the 18th century. And those parish schoolmasters taught local kids. The parents did have to pay a few pennies for their classes, and, true, more parents paid for sons than for daughters. But there were daughters who took classes. And there were no opportunities like this in England at the time. Education was better for women in Scotland. So was their legal position, under common law in Scotland. When the Act of Union was formed in 1707, Scotland retained its own legal system, which had more extensive rights for women than in England.

Q: I know it’s complex, but generally, why was this?

A: Scotland was a much more democratic country, culture, and society than England, period. When Elizabeth I died in 1603, the person who inherited the throne was the King of Scotland James VI, who went to England with his court — which included the Scottish aristocracy. So, the Scottish aristocracy ended up in London. I’m sure they went back to their hunting lodges for the hunting season, but they didn’t live there [in Scotland] and they didn’t set the tone of the country. It was democratized because all that was left were a lot of lawyers and ministers and teachers.

Q: What is distinctive about the ballads in this corpus of songs Anna Gordon knew and documented?

A: A common word about ballads is that there’s a high body count, and they’re all about people dying and killing each other. But that is not true of Anna Gordon’s ballads. They’re about younger women triumphing in the world, often against older women, which is interesting, and even more often against fathers. The ballads are about family discord, inheritance, love, fidelity, lack of fidelity, betrayal. There are ballads about fighting and bloodshed, but not so many. They’re about the human condition. And they have interesting qualities because they’re oral poetry, composed and remembered and changed and transmitted from mouth to ear and not written down. There are repetitions and parallelisms, and other hallmarks of oral poetry. The sort of thing you learned when you read Homer.

Q: So is this a form of culture generated in opposition to those controlling society? Or at least, one that’s popular regardless of what some elites thought?

A: It is in Scotland, because of the enmity between Scotland and England. We’re talking about the period of Great Britain when England is trying to gobble up Scotland and some Scottish folks don’t want that. They want to retain their Scottishness. And the ballad was a Scottish tradition that was not influenced by England. That’s one reason balladry was so important in 18th-century Scotland. Everybody was into balladry partly because it was a unique part of Scottish culture.

Q: To that point, it seems like an unexpected convergence, for the time, to see a more middle-class woman like Anna Gordon transmitting ballads that had often been created and sung by people of all classes.

A: Yes. At first I thought I was just working on a biography of Anna Gordon. But it’s fascinating how the culture was transmitted, how intellectually rich that society was, how much there is to examine in Scottish culture and society of the 18th century. Today people may watch “Outlander,” but they still wouldn’t know anything about this!

Nature Climate Change, Published online: 06 November 2025; doi:10.1038/s41558-025-02472-1

Effective communication of uncertainty is vital for public accurate understanding of climate science. Here the authors find that projections using positive probability terms (for example, a small probability) are perceived as more scientific and trustworthy than those using negative terms (for example, unlikely).

A new 3D human brain tissue platform developed by MIT researchers is the first to integrate all major brain cell types, including neurons, glial cells, and the vasculature, into a single culture. 

Grown from individual donors’ induced pluripotent stem cells, these models — dubbed Multicellular Integrated Brains (miBrains) — replicate key features and functions of human brain tissue, are readily customizable through gene editing, and can be produced in quantities that support large-scale research.

Although each unit is smaller than a dime, miBrains may be worth a great deal to researchers and drug developers who need more complex living lab models to better understand brain biology and treat diseases.

“The miBrain is the only in vitro system that contains all six major cell types that are present in the human brain,” says Li-Huei Tsai, Picower Professor, director of The Picower Institute for Learning and Memory, and a senior author of the open-access study describing miBrains, published Oct. 17 in the Proceedings of the National Academy of Sciences.

“In their first application, miBrains enabled us to discover how one of the most common genetic markers for Alzheimer’s disease alters cells’ interactions to produce pathology,” she adds.

Tsai’s co-senior authors are Robert Langer, David H. Koch (1962) Institute Professor, and Joel Blanchard, associate professor in the Icahn School of Medicine at Mt. Sinai in New York, and a former Tsai Laboratory postdoc. The study is led by Alice Stanton, former postdoc in the Langer and Tsai labs and now assistant professor at Harvard Medical School and Massachusetts General Hospital, and Adele Bubnys, a former Tsai lab postdoc and current senior scientist at Arbor Biotechnologies.

Benefits from two kinds of models

The more closely a model recapitulates the brain’s complexity, the better suited it is for extrapolating how human biology works and how potential therapies may affect patients. In the brain, neurons interact with each other and with various helper cells, all of which are arranged in a three-dimensional tissue environment that includes blood vessels and other components. All of these interactions are necessary for health, and any of them can contribute to disease.

Simple cultures of just one or a few cell types can be created in quantity relatively easily and quickly, but they cannot tell researchers about the myriad interactions that are essential to understanding health or disease. Animal models embody the brain’s complexity, but can be difficult and expensive to maintain, slow to yield results, and different enough from humans to yield occasionally divergent results.

MiBrains combine advantages from each type of model, retaining much of the accessibility and speed of lab-cultured cell lines while allowing researchers to obtain results that more closely reflect the complex biology of human brain tissue. Moreover, they are derived from individual patients, making them personalized to an individual’s genome. In the model, the six cell types self-assemble into functioning units, including blood vessels, immune defenses, and nerve signal conduction, among other features. Researchers ensured that miBrains also possess a blood-brain-barrier capable of gatekeeping which substances may enter the brain, including most traditional drugs.

“The miBrain is very exciting as a scientific achievement,” says Langer. “Recent trends toward minimizing the use of animal models in drug development could make systems like this one increasingly important tools for discovering and developing new human drug targets.”

Two ideal blends for functional brain models

Designing a model integrating so many cell types presented challenges that required many years to overcome. Among the most crucial was identifying a substrate able to provide physical structure for cells and support their viability. The research team drew inspiration from the environment that surrounds cells in natural tissue, the extracellular matrix (ECM). The miBrain’s hydrogel-based “neuromatrix” mimics the brain’s ECM with a custom blend of polysaccharides, proteoglycans, and basement membrane that provide a scaffold for all the brain’s major cell types while promoting the development of functional neurons.

A second blend would also prove critical: the proportion of cells that would result in functional neurovascular units. The actual ratios of cell types have been a matter of debate for the last several decades, with even the more advanced methodologies providing only rough brushstrokes for guidance, for example 45-75 percent for oligodendroglia of all cells or 19-40 percent for astrocytes.

The researchers developed the six cell types from patient-donated induced pluripotent stem cells, verifying that each cultured cell type closely recreated naturally-occurring brain cells. Then, the team experimentally iterated until they hit on a balance of cell types that resulted in functional, properly structured neurovascular units. This laborious process would turn out to be an advantageous feature of miBrains: because cell types are cultured separately, they can each be genetically edited so that the resulting model is tailored to replicate specific health and disease states.

“Its highly modular design sets the miBrain apart, offering precise control over cellular inputs, genetic backgrounds, and sensors — useful features for applications such as disease modeling and drug testing,” says Stanton.

Alzheimer’s discovery using miBrain

To test miBrain’s capabilities, the researchers embarked on a study of the gene variant APOE4, which is the strongest genetic predictor for the development of Alzheimer’s disease. Although one brain cell type, astrocytes, are known to be a primary producer of the APOE protein, the role that astrocytes carrying the APOE4 variant play in disease pathology is poorly understood.

MiBrains were well-suited to the task for two reasons. First of all, they integrate astrocytes with the brain’s other cell types, so that their natural interactions with other cells can be mimicked. Second, because the platform allowed the team to integrate cell types individually, APOE4 astrocytes could be studied in cultures where all other cell types carried APOE3, a gene variant that does not increase Alzheimer’s risk. This enabled the researchers to isolate the contribution APOE4 astrocytes make to pathology.

In one experiment, the researchers examined APOE4 astrocytes cultured alone, versus ones in APOE4 miBrains. They found that only in the miBrains did the astrocytes express many measures of immune reactivity associated with Alzheimer’s disease, suggesting the multicellular environment contributes to that state.

The researchers also tracked the Alzheimer’s-associated proteins amyloid and phosphorylated tau, and found all-APOE4 miBrains accumulated them, whereas all-APOE3 miBrains did not, as expected. However, in APOE3 miBrains with APOE4 astrocytes, they found that APOE4 miBrains still exhibited amyloid and tau accumulation.

Then the team dug deeper into how APOE4 astrocytes’ interactions with other cell types might lead to their contribution to disease pathology. Prior studies have implicated molecular cross-talk with the brain’s microglia immune cells. Notably, when the researchers cultured APOE4 miBrains without microglia, their production of phosphorylated tau was significantly reduced. When the researchers dosed APOE4 miBrains with culture media from astrocytes and microglia combined, phosphorylated tau increased, whereas when they dosed them with media from cultures of astrocytes or microglia alone, the tau production did not increase. The results therefore provided new evidence that molecular cross-talk between microglia and astrocytes is indeed required for phosphorylated tau pathology.

In the future, the research team plans to add new features to miBrains to more closely model characteristics of working brains, such as leveraging microfluidics to add flow through blood vessels, or single-cell RNA sequencing methods to improve profiling of neurons.

Researchers expect that miBrains could advance research discoveries and treatment modalities for Alzheimer’s disease and beyond. 

“Given its sophistication and modularity, there are limitless future directions,” says Stanton. “Among them, we would like to harness it to gain new insights into disease targets, advanced readouts of therapeutic efficacy, and optimization of drug delivery vehicles.”

“I’m most excited by the possibility to create individualized miBrains for different individuals,” adds Tsai. “This promises to pave the way for developing personalized medicine.”

Funding for the study came from the BT Charitable Foundation, Freedom Together Foundation, the Robert A. and Renee E. Belfer Family, Lester A. Gimpelson, Eduardo Eurnekian, Kathleen and Miguel Octavio, David B. Emmes, the Halis Family, the Picower Institute, and an anonymous donor.

A large-scale screen of tuberculosis proteins has revealed several possible antigens that could be developed as a new vaccine for TB, the world’s deadliest infectious disease.

In the new study, a team of MIT biological engineers was able to identify a handful of immunogenic peptides, out of more than 4,000 bacterial proteins, that appear to stimulate a strong response from a type of T cells responsible for orchestrating immune cells’ response to infection.

There is currently only one vaccine for tuberculosis, known as BCG, which is a weakened version of a bacterium that causes TB in cows. This vaccine is widely administered in some parts of the world, but it poorly protects adults against pulmonary TB. Worldwide, tuberculosis kills more than 1 million people every year.

“There’s still a huge TB burden globally that we’d like to make an impact on,” says Bryan Bryson, an associate professor of biological engineering at MIT and a member of the Ragon Institute of Mass General Brigham, MIT, and Harvard. “What we’ve tried to do in this initial TB vaccine is focus on antigens that we saw frequently in our screen and also appear to stimulate a response in T cells from people with prior TB infection.”

Bryson and Forest White, the Ned C. and Janet C. Rice Professor of Biological Engineering at MIT, and a member of the Koch Institute for Integrative Cancer Research, are the senior authors of the study, which appears today in Science Translational Medicine. Owen Leddy PhD ’25 is the paper’s lead author.

Identifying vaccine targets

Since the BCG vaccine was developed more than 100 years ago, no other TB vaccines have been approved for use. Mycobacterium tuberculosis produces more than 4,000 proteins, which makes it a daunting challenge to pick out proteins that might elicit a strong immune response if used as a vaccine.

In the new study, Bryson and his students set out to narrow the field of candidates by identifying TB proteins presented on the surface of infected human cells. When an immune cell such as a phagocyte is infected with Mycobacterium tuberculosis, some of the bacterial proteins get chopped into fragments called peptides, which are then displayed on the surface of the cell by MHC proteins. These MHC-peptide complexes act as a signal that can activate T cells.

MHCs, or major histocompatibility complexes, come in two types known as class I and class II. Class I MHCs activate killer T cells, while class II MHCs stimulate helper T cells. In human cells, there are three genes that can encode MHC-II proteins, and each of these comes in hundreds of variants. This means that any two people can have a very different repertoire of MHC-II molecules, which present different antigens.

“Instead of looking at all of those 4,000 TB proteins, we wanted to ask which of those proteins from TB actually end up being displayed to the rest of the immune system via MHC,” Bryson says. “If we could just answer that question, then we could design vaccines to match that.”

To try to answer the question, the researchers infected human phagocytes with Mycobacterium tuberculosis. After three days, they extracted MHC-peptide complexes from the cell surfaces, then identified the peptides using mass spectrometry.

Focusing on peptides bound to MHC-II, the researchers found 27 TB peptides, from 13 proteins, that appeared most often in the infected cells. Then, they further tested those peptides by exposing them to T cells donated by people who had previously been infected with TB.

They found that 24 of these peptides did elicit a T cell response in at least some of the samples. None of the proteins from which these peptides came worked for every single donor, but Bryson believes that a vaccine using a combination of these peptides would likely work for most people.

“In a perfect world, if you were trying to design a vaccine, you would pick one protein and that protein would be presented across every donor. It should work for every person,” Bryson says. “However, using our measurements, we’ve not yet found a TB protein that covers every donor we’ve analyzed thus far.”

Enter mRNA vaccines

Among the vaccine candidates that the researchers identified are several peptides from a class of proteins called type 7 secretion systems (T7SSs). Some of these peptides also turned up in an earlier study from Bryson’s lab on MHC-1.

“Type 7 secretion system substrates are a very small sliver of the overall TB proteome, but when you look at MHC class I or MHC class II, it seems as though the cells are preferentially presenting these,” Bryson says.

Two of the best-known of these proteins, EsxA and EsxB, are secreted by bacteria to help them escape from the membranes that phagocytes use to envelop them within the cell. Neither protein can break through the membrane on its own, but when joined together to form a heterodimer, they can poke holes, which also allow other T7SS proteins to escape.

To evaluate whether the proteins they identified could make a good vaccine, the researchers created mRNA vaccines encoding two protein sequences — EsxB and EsxG. The researchers designed several versions of the vaccine, which were targeted to different compartments within the cells.

The researchers then delivered this vaccine into human phagocytes, where they found that vaccines that targeted cell lysosomes — organelles that break down molecules — were the most effective. These vaccines induced 1,000 times more MHC presentation of TB peptides than any of the others.

They later found that the presentation was even higher if they added EsxA to the vaccine, because it allows the formation of the heterodimers that can poke through the lysosomal membrane.

The researchers currently have a mix of eight proteins that they believe could offer protection against TB for most people, but they are continuing to test the combination with blood samples from people around the world. They also hope to run additional studies to explore how much protection this vaccine offers in animal models. Tests in humans are likely several years away.

The research was funded by the MIT Center for Precision Cancer Research at the Koch Institute, the National Institutes of Health, the National Institute of Environmental Health Sciences, and the Frederick National Laboratory for Cancer Research.

A robot searching for workers trapped in a partially collapsed mine shaft must rapidly generate a map of the scene and identify its location within that scene as it navigates the treacherous terrain.

Researchers have recently started building powerful machine-learning models to perform this complex task using only images from the robot’s onboard cameras, but even the best models can only process a few images at a time. In a real-world disaster where every second counts, a search-and-rescue robot would need to quickly traverse large areas and process thousands of images to complete its mission.

To overcome this problem, MIT researchers drew on ideas from both recent artificial intelligence vision models and classical computer vision to develop a new system that can process an arbitrary number of images. Their system accurately generates 3D maps of complicated scenes like a crowded office corridor in a matter of seconds. 

The AI-driven system incrementally creates and aligns smaller submaps of the scene, which it stitches together to reconstruct a full 3D map while estimating the robot’s position in real-time.

Unlike many other approaches, their technique does not require calibrated cameras or an expert to tune a complex system implementation. The simpler nature of their approach, coupled with the speed and quality of the 3D reconstructions, would make it easier to scale up for real-world applications.

Beyond helping search-and-rescue robots navigate, this method could be used to make extended reality applications for wearable devices like VR headsets or enable industrial robots to quickly find and move goods inside a warehouse.

“For robots to accomplish increasingly complex tasks, they need much more complex map representations of the world around them. But at the same time, we don’t want to make it harder to implement these maps in practice. We’ve shown that it is possible to generate an accurate 3D reconstruction in a matter of seconds with a tool that works out of the box,” says Dominic Maggio, an MIT graduate student and lead author of a paper on this method.

Maggio is joined on the paper by postdoc Hyungtae Lim and senior author Luca Carlone, associate professor in MIT’s Department of Aeronautics and Astronautics (AeroAstro), principal investigator in the Laboratory for Information and Decision Systems (LIDS), and director of the MIT SPARK Laboratory. The research will be presented at the Conference on Neural Information Processing Systems.

Mapping out a solution

For years, researchers have been grappling with an essential element of robotic navigation called simultaneous localization and mapping (SLAM). In SLAM, a robot recreates a map of its environment while orienting itself within the space.

Traditional optimization methods for this task tend to fail in challenging scenes, or they require the robot’s onboard cameras to be calibrated beforehand. To avoid these pitfalls, researchers train machine-learning models to learn this task from data.

While they are simpler to implement, even the best models can only process about 60 camera images at a time, making them infeasible for applications where a robot needs to move quickly through a varied environment while processing thousands of images.

To solve this problem, the MIT researchers designed a system that generates smaller submaps of the scene instead of the entire map. Their method “glues” these submaps together into one overall 3D reconstruction. The model is still only processing a few images at a time, but the system can recreate larger scenes much faster by stitching smaller submaps together.

“This seemed like a very simple solution, but when I first tried it, I was surprised that it didn’t work that well,” Maggio says.

Searching for an explanation, he dug into computer vision research papers from the 1980s and 1990s. Through this analysis, Maggio realized that errors in the way the machine-learning models process images made aligning submaps a more complex problem.

Traditional methods align submaps by applying rotations and translations until they line up. But these new models can introduce some ambiguity into the submaps, which makes them harder to align. For instance, a 3D submap of a one side of a room might have walls that are slightly bent or stretched. Simply rotating and translating these deformed submaps to align them doesn’t work.

“We need to make sure all the submaps are deformed in a consistent way so we can align them well with each other,” Carlone explains.

A more flexible approach

Borrowing ideas from classical computer vision, the researchers developed a more flexible, mathematical technique that can represent all the deformations in these submaps. By applying mathematical transformations to each submap, this more flexible method can align them in a way that addresses the ambiguity.

Based on input images, the system outputs a 3D reconstruction of the scene and estimates of the camera locations, which the robot would use to localize itself in the space.

“Once Dominic had the intuition to bridge these two worlds — learning-based approaches and traditional optimization methods — the implementation was fairly straightforward,” Carlone says. “Coming up with something this effective and simple has potential for a lot of applications.

Their system performed faster with less reconstruction error than other methods, without requiring special cameras or additional tools to process data. The researchers generated close-to-real-time 3D reconstructions of complex scenes like the inside of the MIT Chapel using only short videos captured on a cell phone.

The average error in these 3D reconstructions was less than 5 centimeters.

In the future, the researchers want to make their method more reliable for especially complicated scenes and work toward implementing it on real robots in challenging settings.

“Knowing about traditional geometry pays off. If you understand deeply what is going on in the model, you can get much better results and make things much more scalable,” Carlone says.

This work is supported, in part, by the U.S. National Science Foundation, U.S. Office of Naval Research, and the National Research Foundation of Korea. Carlone, currently on sabbatical as an Amazon Scholar, completed this work before he joined Amazon.

For many in the research community, it’s gotten harder to be optimistic about the impacts of artificial intelligence.

As authoritarianism is rising around the world, AI-generated “slop” is overwhelming legitimate media, while AI-generated deepfakes are spreading misinformation and parroting extremist messages. AI is making warfare more precise and deadly amidst intransigent conflicts. AI companies are exploiting people in the global South who work as data labelers, and profiting from content creators worldwide by using their work without license or compensation. The industry is also affecting an already-roiling climate with its ...

Ten years after nations adopted the Paris Agreement, its objectives are in jeopardy amid rising climate pollution and a political backlash against clean energy.