Inspired by networks found in nature, engineers design stronger, more resilient “metamaterials”

Order has long been the rule in the design of synthetic “metamaterials,” which have internal structures that give the whole material exceptional properties not found in nature.

When scientists engineer a metamaterial, they design from a single building block, or “unit cell,” that is then copied and pasted many times over. The result is a precise, periodic microstructure that results in improved material performance. 

Now, a team from MIT and Northeastern University has found that engineering a little disorder into a material can further enhance its “meta” features. 

In an open-access study appearing today in the journal Nature, the researchers introduce a new way to design metamaterials, inspired by the types of networks that appear in a leaf’s vasculature, the veins in a dragonfly’s wings, and the virtual connections in a social network. These new metamaterials — termed “scale-rich metamaterials” — exhibit structures of varying sizes, which sprawl and connect to form a physical network.

For decades, networks have been used as theoretical constructs to describe everything from cities to the internet to connections between neurons. The team has now applied such network designs to physical materials. They designed metamaterials made not from repeating unit cells, but from tiny support beams of varying sizes, similar to a network’s dense, web-like patterns. 

When subjected to physical tests, the new materials attained a widely controllable range of stiffness and held up against sudden impacts better than conventional, ordered metamaterials. They also were able to cushion and protect fragile objects embedded within the materials.

The team designed their materials using an algorithm which generates an optimal network of microstructures for a desired material property. The procedure can be used to design resilient, impact-absorbing materials such as helmets and armor, as well as acoustic lenses — materials whose microstructure efficiently focuses sound or vibrations.

“This is a completely new way of thinking for metamaterials,” says study author Carlos Portela, associate professor of mechanical engineering at MIT. “Unlocking this network design for materials really frees us to think of new microstructures and mechanical benefits that haven’t been possible before.” 

The study’s MIT co-authors include Andrew Yen-Jong Chen, Mohammad Charara, and Jayna Wadhwa, along with Csaba Both, Ting-Ting Gao, and Albert-László Barabási of Northeastern University, and Niek Mooij of Utrecht University in the Netherlands.

Looking beyond order

Network science is a field of research that studies complex networks, the nodes and connections within a network, and how these elements relate to each other and to the network as a whole. Network science has been used to map and characterize a wide range of systems, from networks in nature, such as food webs, evolutionary trees, and neural connections, to financial markets, cellular and social networks, and the internet. 

Albert-László Barabási, a co-author of the new study, has been at the forefront of network science. A professor of physics at Northeastern University, Barabási explores the structure of complex networks across many disciplines, studying where a network may exhibit vulnerabilities and how it could be made more robust.

Most of Barabási’s work has involved conceptual networks, or networks that are mapped onto existing systems. Recently, he wondered: Could network science be applied to build a physical material with certain desired properties that are not possible in nature? 

For this, Barabási joined efforts with Portela, whose group at MIT specializes in metamaterials. The researchers engineer materials from the ground up, with precisely patterned microstructures that enhance the material’s performance overall. Like others in the field, Portela has built metamaterials from a single unit cell, repeating this building block in patterns such as square lattice structures and honeycomb architectures.

In many cases, Portela has found that such ordered patterns do improve a material’s strength and resilience. But if these materials are punctured or cracked, the rest quickly buckles and fails. He says that’s because the internal structure is all the same; if one region of the material is deformed, the surrounding, similar unit cells will respond and deform in kind.

“We’ve spent years looking at periodic and ordered microstructures, and we’ve learned there are a lot of limitations and non-ideal responses, such as this localization effect,” Portela says. “We wanted to break free of that notion and use network science to design these materials.”

Line by line

The team developed an algorithm to virtually generate a network of beams with the constraints of a physical material. When applying network science to something like the internet, researchers haven’t had to consider physical constraints, such as limiting the number, density, or intersection of connections. In a physical material, the number and thickness of beams, and the space between them, start to matter. 

To design a physical network, the team’s algorithm starts by drawing a point at a random coordinate in a domain representing the size of the final material. The program draws a line at a random angle, which stretches through the point and crosses the entire space. The program then repeats the process with a second point, at a second random coordinate, and a line drawn through the point at another random angle, until that second line reaches a boundary, of either the whole domain, or the first drawn line. The process then repeats, again and again, drawing progressively shorter, thinner lines. The resulting material represents a dense network of intersecting beams of many varying sizes. 

Scientists can vary the type of network that the program generates, by varying the thickness of the starting line and how fast that thickness decreases with each successive line drawn. By simply tuning these two parameters, the team found they could generate a huge range of networks. They then carried out simulations to predict how tuning one or the other, or both, by varying degrees, would change a material’s stiffness and other physical properties. 

“We see we can really tune the stiffness of a material just by changing these two parameters, which dictate how these network patterns get generated,” Portela says. 

The team then fabricated physical materials with microstructures that were patterned after the networks they virtually generated. They used 3D-printing methods to deposit polymer into patterns of tiny, interconnecting beams. Each resulting sample was a large square, about the size of a credit card, with internal beams that ranged from microns to centimeters in scale. 

They put the samples through various physical tests, such as dropping a large weight onto them and squeezing them with increasing force. They recorded the materials’ response, along with samples of conventional lattice and honeycomb metamaterials. The network-inspired samples, they found, were more resilient to compression and impacts. 

“Because there is such a richness of length scales in these networked materials, if one part fails, there is a strong web of structures around it that keeps the entire material from failing catastrophically,” Portela says. “It’s more efficient at distributing damage and absorbing energy.”

As another demonstration, the researchers designed a material inspired by the density-graded microstructure of a pomelo’s thick rind. They designed the material with edges made from a denser network of beams, and showed it could cushion and protect a fragile object that was embedded at the center of the network.

“This work provides a design framework for materials whose features vary over a very wide range,” says materials expert Tian “Tim” Chen, associate professor in the Department of Aeronautics at Imperial College London, who was not involved with the study. “That may lead to a next generation of materials with unprecedented physical behavior, for any structure that must carry load, absorb energy or control vibration.”

This work was supported, in part, by the U.S. National Science Foundation’s Designing Materials to Revolutionize and Engineer our Future (DMREF) program.

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

Streamlining metal parts production to revitalize American industry

The United States used to be the world leader in metal casting, a process where liquid hot metal is poured into a mold to solidify into a desired shape. But more than 70 percent of American metal casting facilities have closed since the industry’s peak in the 1950s. Today, China and India dominate global metal casting production, which is critical for making the components of jet engines, nuclear and gas turbines, medical devices, and more.

Fabri is trying to revitalize American metal casting. The company, founded by Steven Davis ’24; Thomas Cole ’91, MBA ’13; and Pieter Coulier, is building a new kind of metal casting foundry that uses additive manufacturing, software, and vertical integration to eliminate many of the most labor-intensive steps associated with traditional metal casting. 

Fabri is already producing parts for a growing list of customers at its 16,000-square-foot plant in Billerica, Massachusetts, where it is applying its approach to one of the most multidisciplinary and challenging types of metal casting, known as investment casting. The company is working with Raytheon, Lockheed Martin, and Sikorsky Aircraft, among others, and is moving through qualification to make parts for the most critical aerospace and defense applications.

Fabri’s first foundry was designed to be quickly replicated. Even as it significantly expands capacity at its current plant, demand from customers is driving the company to build a far larger facility in 2027 that will produce bigger parts at higher volume as customer programs move into production.

“Metal casting is one of the oldest and most prevalent manufacturing methods in the world,” Davis says. “It’s a uniquely difficult process because there are so many different steps. There’s so much domain knowledge and experience you need to understand it. If we get cut off from the world’s casting capacity, that means no more pots and pans, no more storm drains and other basic things, but also no more aircraft. Lots of medical devices, like titanium hip implants, are cast. Power generation also depends on metal parts. These are very critical industries that we need to continue to produce.”

Faster metal production

Davis grew up competing in robotics competitions like BattleBots. The biggest challenge for his small, underfunded teams was getting the parts to make the robots.

Davis arrived at MIT in 2020, where he majored in mechanical engineering. As part of a summer internship after his sophomore year, he worked at Relativity Space, which builds reusable rockets. There he saw multi-million-dollar projects being delayed by the same hurdles he faced in robotics competitions. His team would wait weeks for expensive machines to make parts.

“I realized the problem was scale,” Davis says. “These production processes couldn’t even scale to aerospace, so they weren’t going to scale to far larger industries.”

When Davis returned to MIT that fall, he worked on adapting the precision and flexibility of 3D printing to producing metal parts. Davis also took several classes taught by John Hart, MIT’s Class of 1922 Professor and head of the Department of Mechanical Engineering. Hart co-founded several additive manufacturing companies including VulcanForms and Desktop Metal. He continues to advise Fabri.

“I was very interested in finding the best uses of additive manufacturing,” Davis recalls. “A lot of the most successful companies weren’t making end parts; they were making tools. My favorite example is Invisalign. They’re not printing the dental liners: They are printing the molds that make the aligners.”

Davis began thinking about starting a company roughly midway through his bachelor’s degree at MIT. He met Cole through his MIT network. Cole had founded a fully automated CNC machining company making dental implants that was acquired by AstraZeneca. Pieter Coulier, Fabri’s third co-founder, previously developed the simulation and metal 3D-printing algorithms at Velo3D that enabled SpaceX’s Raptor engines to be printable.

Davis also began visiting foundries, where he repeatedly heard their biggest bottleneck was labor.

“They all have huge ‘Help Wanted’ signs out front,” Davis says. “Capacity is going down. That’s why we realized automation had to be a core part of our thesis. According to the American Foundry Society, these workers are retiring at a rate of 3-to-1. It’s really hard to find these foundry engineers who understand what goes into the process and can design these molds. We have customers tell us they had an engineer retire and now they can’t make certain parts anymore. We want everything we learn to be encoded into our software forever.”

Investment casting is a complex, multistep process.

“Traditional investment casting starts by cutting a hard tool out of steel,” Davis explains. “Then you injection mold wax patterns into it. Those wax parts get glued together into what’s known as a tree. Then you dip the tree into liquid loaded with ceramic powder, you stucco it with sand, dry it, and repeat the process seven or eight times. That’s how you get the ceramic mold to pour your steel into to make the part.”

Fabri’s process starts with the part’s computer-aided design (CAD) file. The company’s proprietary software, FoundryOS, designs a pattern for each part, which Fabri prints in resin using established techniques including digital light processing (DLP) and stereolithography. The printed pattern is dipped in ceramic slurry to build the shell that molten metal is poured into. FoundryOS incorporates data from every step and uses machine learning models to improve the process and the parts with each casting.

“By printing that mold right off the bat, we remove the need for all those early steps,” Davis explains. “We fully automate the creation of the [complex patterns and channels] needed for making that tree. Instead, we print it all in one piece and consolidate a lot of that work.”

Davis says traditional tooling can cost upward of $1 million for complex enough designs and take many months to make — a big reason companies have to wait so long for parts.

Davis built the first prototype of Fabri’s printer in the basement of his MIT fraternity using funds from the MIT Sandbox Innovation Fund. After subleasing a space in Ashland, Massachusetts, Fabri was pouring metal within six months. Today, he says Fabri’s approach also cuts post processing and benefits from the increased flexibility of 3D printing, which allows Fabri to iterate on parts in days.

“What really hurts traditional foundries is they put a design into production, they tool up, they cut a tool, and when they get their first part, they realize something’s wrong,” Davis says. “Then they have to rework it or fix it. They’d like to fix the tooling in the process, but they can’t because they spent so much on the tool and waited six to nine months for it.”

Reinvigorating an industry

Fabri opened its first foundry in Ashland in 2025 and moved production to Billerica in spring of 2026. Davis says the company can go from receiving a design to producing a finished part in as little as a week. Fabri is making parts out of aluminum, copper alloys, and stainless steel, primarily for munitions, aviation, and naval applications. In early 2027, the company plans to begin casting nickel superalloys, the metals used in the turbine blades that can currently have four-year wait times.

The company’s process also enables complex geometries and organic topologies that hard tooling can’t produce.

“We’re building the software that can make these parts more intelligently and faster,” Davis says. “Qualifying for aerospace production with a company like Raytheon is a really big deal. Most startups and 3D-printing companies we know haven’t done it. The ability for us to go from opening our first facility in 2025 to now going into production with some of the nation’s most critical platforms is a huge milestone. We are going to be making parts that are going to fly in the systems and go into the munitions platforms that the country needs.”

de MIT News https://ift.tt/z2HjXUE

In postcolonial Ghana, a fresh start but not a blank slate

It was a landmark in history: In 1957, Ghana became the first sub-Saharan African state to achieve independence from colonial rule, having previously been under the control of Great Britain. 

But gaining political independence does not mean the past is left behind, and there has long been public debate about the many long-term effects of the colonial period on state and society in subsequent decades.  

Now, an empirically rich study explores one aspect of the Ghanaian state, the composition of its bureaucracy, during and after British rule. The study finds that two ethnic groups, the Ga and the Fante, which made up 14 percent of the population combined, held over 70 percent of public-sector positions in the 1950s, suggesting goods such as desirable jobs had been heavily distributed along group lines. 

“At the moment of independence, you have a state that is not meritocratic, and is highly captured by certain ethnic groups,” says MIT scholar Noah Nathan, co-author of a newly published paper detailing the study’s findings. As a result, he adds, “You can imagine the politics that creates, which are demands to equalize access to state power and state authority. And things start off on this foot of ethnic political competition over the state. That is in part a legacy of colonial state-building.”

At least in this case, “Independence in Africa wasn’t this hard break from the past,” suggests Nathan, a Ford International Professor of Political Science in MIT’s Department of Political Science. “Africans became the leaders of their countries, but the state institutions that they inherited weren’t born fresh at independence. They were carried over from the colonial period.”

The paper, “The Seeds of State Capture: Merit and Patrimonialism in the Colonial Bureaucracy,” is published in advance online form in the American Political Science Review. Nathan’s co-authors are Sarah Brierley, an associate professor at the London School of Economics and Political Science; George Kwaku Ofoso, an associate professor at the London School of Economics and Political Science; and Tingxuan Zhu PhD ’26, a recent graduate of MIT’s doctoral program in political science at MIT. 

Analyzing names

To conduct the study, the scholars dug into records from the period of British rule in Ghana, which extended from the 19th century until 1957. Using documents from the British National Archives, they examined individual-level employment records for 11,702 bureaucrats from the colonial-era civil service, building a dataset spanning from 1897 to 1955. The records included 6,879 Ghanaian civil servants and 4,805 European civil servants. The scholars’ aim was to shed light on the dynamics of civil service hiring and politics. 

British leaders at the time, and some commentators subsequently, have cast the colonial state as one built on rational, merit-based principles — what scholars call a “Weberian” bureaucracy, named after the famous German sociologist Max Weber. Scholars say that other types of bureaucracies, in contrast, display “patrimonialism,” with jobs being assigned not on merit but through patronage.

To analyze this question further, the scholars looked at 13.8 million names of registered voters from contemporary Ghana, along with localized ethnicity data from the 2010 national census. They used this to establish a baseline of the relationship between names and ethnicity in the country. Then, the scholars went back to the historical archives, and, using the names there, determined the ethnicities of past Ghanaian civil servants, from 1897 to 1955. 

Ultimately this approach showed that on the eve of independence, the Ga and Fante had managed to gain a percentage of civil service jobs that was five times greater than the percentage of the population they represented. That strongly suggests that the colonial-era bureaucracy built under British rule was not merit-based, at least in terms of overall composition. 

“Scholars have said this post-1957 independent Ghana brings with it the introduction of patrimonialism into the bureaucracy, and we’re saying this was going on long before,” Nathan explains. “It actually wasn’t meritocratic at all, and Kwame Nkrumah, the first president of Ghana, didn’t inherit some Weberian rational bureaucracy that he then destroyed. He inherited this thing rife with patrimonialism and patronage, that he then adapted. And that’s a very different interpretation of the post-independence period.”

Exams and education

To further examine the question of merit, the scholars even obtained individual-level results from Ghana’s civil service exams, then analyzed those in relation to which applicants got hired — or passed over for jobs. The results indicate little correspondence between exam results and job offers, again suggesting that ethnicity and social ties mattered more in hiring, on aggregate. 

“If you were from the right ethnic group, you were going to get a job, even if you did pretty poorly on the civil service exam,” Nathan says. “And people from disfavored ethnic groups who did much better on the exam were not getting the same positions.”

The study also rules out a seemingly logical explanation for at least part of the ethnic skew in the composition of the Ghanaian civil service — that hiring reflected levels of educational attainment, even if access to schooling itself might have reflected ethnic iniquities. But the data indicate that was not a factor. 

The scholars located every school in the country geographically, dating to the 1890s, and closely measured access to education for all ethnic groups over time. The Ga and Fante groups may have had some initial advantages in this regard, but by the 1930s, many other groups had caught up in educational attainment and still were not procuring the same relative numbers of civil services jobs through the 1950s. 

“At least in this case, that can’t explain why these two groups got favored in the hiring process,” Nathan says. “It wasn’t just a direct outcome of the fact that some parts of society had more education than others.”

As it happens, geographical proximity to civil service jobs is also not an explanatory factor; the scholars also examined the locations of government offices and found they cannot account for the quantity of jobs going to the Ga and Fante. 

Out of sight

Instead, the researchers believe the likeliest scenario is that, once established in the civil service, people from the Ga and Fante managed to use their social connections to get more and more jobs offered to others from their own groups. Moreover, they may have done so not necessarily with explicit British approval, but with something more resembling indifference on the part of the colonial rulers. 

“The mechanics of the hiring process were very decentralized,” Nathan says. “And the British didn’t know very much about the society they were governing. That created an information asymmetry where local bureaucrats could get friends into jobs in a way their superiors were either not aware of or didn’t have very much incentive to stop.”

All told, Nathan observes, “The ministry of finance in Ghana in 1957 isn’t some new thing that’s created from scratch. It’s the colonial ministry of finance that’s been in place for 60 years, and the bureaucrats in it showed up to work the next day in the same jobs. [The new leaders] inherited thousands of employees who had gotten there through this process.”

The study is one of several that Nathan has published using data to answer questions that span both political science and history. Quantifying the issue in a granular way helps illuminate the large-scale political dynamics.

“People have written about this in broad strokes, even the fact that certain ethnic groups in some countries got more access to positions than others, but they haven’t had detailed microdata on exactly who got which job and when,” Nathan says. “That’s the innovation of our paper … we could unpack the nuts and bolts of how this hiring process happened, and in doing so dispel some conventional wisdom about how African states were built in the colonial period.”

The new paper is not the only piece of scholarship taking this kind of data-intensive approach to studying colonial state-building and its legacy. Ada Johnson-Kanu, an assistant professor of political science at the University of Kentucky, has conducted this sort of study of Nigeria’s bureaucracy over time, in work she completed in 2021. 

Nathan, for his part, says he hopes that other scholars can expand upon this kind of data-driven approach to digging into certain historical questions. 

“There is just incredibly rich archival evidence out there in the world for a lot of countries,” Nathan observes. “I think this is a really fun way to do research.”

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

AI-assisted learning platform helps robots reason beyond literal commands

Humans are intuitive by nature. We can sense meaning beyond the literal words that we encounter. Now, a new artificial intelligence-assisted learning platform developed by MIT engineers is enabling robots to read larger meaning into a human’s natural language commands. 

In a study to be presented at the annual Conference on Robot Learning this month, researchers from MIT and New York University introduce FLEX, a flexible feature-editing pipeline for robot motion-planning. 

With FLEX, humans can interact with a robot by giving it instructions in natural, conversational language. The robot can not only follow the instructions literally but also read into the words to draw larger generalizations that help to improve its motion-planning task. 

The team demonstrated FLEX in several simulated scenarios, including one in which a robot serves a cup of coffee. In these virtual scenarios, a person interacts with a robotic arm by typing commands in natural language. As the robot moves the coffee cup over a cluttered table, the person might type “Don’t spill on the laptop!” In response, the robot will reason about why it should avoid the laptop. For this, FLEX incorporates a large language model (LLM) — a type of AI system that is trained on enormous libraries of text, including social media posts, video transcriptions, and web content, which the LLM uses to identify larger trends, patterns, and generalizations. 

“LLMs have seen so many texts from the internet,” says study author Nathan Dennler, a postdoc in MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). “People might have written about, ‘Oh, my phone dropped in the water and now it doesn’t work,’ or ‘I spilled coffee on my computer and now it doesn’t work.’ So LLMs pick up on the idea that liquids plus devices leads to things not working.”

Through the LLM, the robot reasons that it should also avoid bumping into all electronics. In its updated path, the robot adjusts to avoid both the laptop and a nearby phone, without the person having to explicitly spell this out.

“The idea is that there is some higher-level intention behind the words that you say,” Dennler says. “People naturally get this. But you have to carefully explain that to a robot, which takes time and effort. FLEX gives you an interface of communication with a robot to say, ‘When people say this, they mean that.’”

The study was led by Andreea Bobu, assistant professor of aeronautics and astronautics at MIT. Co-authors are Audrey Lee SM ’26 and Ilia Sucholutsky of New York University.

Beyond what we say

At MIT, Bobu heads up the Collaborative Learning and Autonomy Research (CLEAR) Lab, where her group develops autonomous agents that learn to work for, with, and around people. She is focused on developing assistive and household robots that can carry out physical tasks, and reason and adapt in real-time, in response to a person’s natural, conversational commands. 

“Robots use math to do their trajectory planning. But we can’t expect a consumer to learn to do this math in order to tell a robot what to do,” says Dennler, who is a member of the CLEAR Lab. “Instead of having to get people to learn how to use a robot in a specific way, the goal is to interact with the robot through natural language.”

There is a large effort in the robotics community to incorporate natural language into robotic learning pipelines. Various strategies include using language to clarify a task’s goal, redirect a robot’s trajectory, and update its understanding of what is a safe path to take. These existing methods use natural language to tell the robot how to physically perform a task. But a robot could potentially learn much more from what we tell it. 

Bobu and her colleagues are expanding what robots can learn from natural language commands. In their new study, they focused on enabling a robot to interpret a person’s underlying, unspecified preference from limited natural language commands.  

Absorbing common sense

The team’s new robot learning platform, FLEX, is a three-step process that helps a robot read more into a natural language command, beyond what it is explicitly told to do. 

The first step, “Add,” instructs a robot to add a feature to its list of things in its environment that matter for the task at hand, based on a user’s natural language command. For instance, as a robot is moving a bowling ball from one location to another, a person might interject with “Don’t drop it on my mug!” The robot would then add the mug to a list of things to avoid in its path. This first step is how previous methods have worked. Every time a person makes a correction, the robot adds to its features list, and updates its trajectory to avoid those features. 

“The problem happens when your feature list becomes super long, and some features are no longer relevant because your environment may have changed,” says Dennler, who notes that this approach can grow tiresome if a person has to specify every specific feature to avoid. 

The second step, “Generalize,” is the new solution that FLEX provides. This step incorporates an LLM — in this case, GPT-4o — that considers the mug, along with any other features that the person has so far specified to avoid. The LLM automatically searches through its vast cache of data for patterns and connections among these features, to ultimately make a generalization that it should avoid objects that are fragile, even when the person has not explicitly said so. 

“LLMs have seen so many texts from the internet, from all sorts of people and places,” Dennler says. “Because they’ve trained so much on those things, some of that common sense gets absorbed into the LLM.”

In the third step, “Mask,” the robot’s standard computer vision system assesses its environment to determine what objects are no longer present, and then deletes those objects from its features list. This entire three-step process takes about five seconds. 

The team tested FLEX in simulations of a robot arm carrying out three different tasks: serving a cup of coffee, moving a candle, and carrying a bowling ball. In all three scenarios, volunteers interacted with the virtual robot by typing natural language commands, such as “Don’t spill on my laptop,” “Stay away from the stack of paper,” and “Don’t drop on the mug.” With FLEX, the robot followed all three commands literally and also reasoned at a higher level to not spill on electronics, stay away from flammable objects, and not drop its cargo on fragile things. 

“Everyone’s homes are different, and everyone cares about different things, so no robot comes out of the box knowing what matters to you,” Bobu says. “What excites me about FLEX is that it can pick that up from a sentence or two instead of asking you to spell out every rule — and that’s the difference between a robot people put up with and one they actually want in their home.” 

This work was partly supported by the MIT Generative AI Impact Consortium and the MIT Schwarzman College of Computing’s Social and Ethical Responsibilities of Computing program.

de MIT News https://ift.tt/s5pzTyu

How to predict the behavior of molecules from scant data

Discovering new molecules for specific uses is a major driver of progress in a wide variety of industries, whether it’s pharmaceutical companies looking for new drugs, or electronics or energy companies seeking new materials with particular properties. But it’s also largely a trial-and-error process, because there is not yet any way to perfectly predict how a given chemical compound will behave in terms of the myriad different properties and responses materials can exhibit.

Recent progress has been made on using artificial intelligence and machine learning to supplement or speed up laborious calculations based on physical principles that can at least provide some initial guidance on which compounds might be worth synthesizing and testing. Now, a new method developed at MIT shows that when there are few data on some technologically relevant property, such as ligand affinities for a recently discovered cellular target for a potential drug, this approach can outperform other existing methods for winnowing down the field of possible molecules that might be worth looking at.

The findings were reported in the Sept. 14 issue of the Journal of Chemical Information and Modeling, in a paper by MIT graduate student Akshat Shirish Zalte and recent graduate Jackson Burns PhD ’26, Research Scientist Charlles Abreu, Professor William Green, and three researchers from the chemical company BASF. 

An earlier version of the paper has been online since last year, and the open-source model the team developed, called CheMeleon, has already found regular use by several pharmaceutical companies. CheMeleon was built using ChemProp, an open-source machine-learning package also developed and maintained by the Green group at MIT, and anyone can download CheMeleon through ChemProp and fine-tune it for their own purposes.

Earlier this year, there was a contest organized by OpenADMET for trying to predict the properties of a variety of new molecules that had not previously been synthesized, says Green, who is the Hoyt Hottel Professor of Chemical Engineering. “It was a big contest, a lot of people participated,” he says. When the winners were announced, it turned out that nine of the top 20 entries had used CheMeleon in their process. “So then we started getting all kinds of inquiries from the pharma companies about this, but our paper had just been submitted; it wasn’t even peer-reviewed or anything yet.”

While the paper went through that review process, its influence was already spreading. At least one pharma company, Green says, “sent a message to all their people, and said ‘Anytime you’re modeling a property, start with CheMeleon, that’s our new company policy’” when making initial decisions about which molecules to pursue.

The key to CheMeleon was pre-training a machine-learning model with detailed data about a million different molecules, so that it gets to learn correlations between the structure of a given molecule and how it behaves with respect to a catalog of about 1,600 descriptors that have been developed by chemists over decades. CheMeleon can then be fine-tuned to predict well-known properties like solubility and viscosity, to more esoteric properties such as bioactivity or reactivity, with a certain protein or enzyme.

Those million molecules are a minuscule fraction of the innumerable ways that atoms can bond together, but they provide enough examples so that once the AI system has been trained on them, it can work out on its own many underlying patterns and relationships, and then apply those to a new, unknown molecule that’s presented to it.

“Most of today’s deep-learning models require more data than traditional methods,” says Zalte. Some successful models use extensive data from the published literature in chemical journals, he says. “We don’t look at the literature at all. We just calculate a bunch of molecular descriptors like molecular weight or number of oxygens, among others.” Then predicting those descriptors forms the basis for their pre-training task.

CheMeleon is a pre-trained foundation model. Inspired by the well-known foundation models in language such as ChatGPT and Claide, which essentially are developed by feeding them massive amounts of information from the internet, CheMeleon uses molecular descriptors of a million molecules from PubChem, a large, freely available database of chemical information maintained by the U.S. National Institutes of Health. 

With classical expert-driven models, on the other hand, specialists in a field manually input detailed knowledge about their subject. In developing these classical models, he says, for example, people know from experiments that a certain property of interest, say solubility, may relate to particular descriptors of the same molecules, “like molecular weight, or the number of oxygens that are present, or how branchy or bulky the molecule is.” 

With this team’s approach, they can start with an existing neural network architecture and train it to learn underlying principles. “We’re not training it for a specific property, but as a general model that understands some level of chemistry, so that even if you fine-tune it with just a few data points for the task you care about, you can still get some good performance,” he says. That’s thanks to the fact that the model is “not having to learn everything from scratch.”

To verify its usefulness, “We compared our model not only against other foundation models that are out there, but also with classical models leveraging expert-designed principles,” Zalte says. Comparing its performance across 58 test cases, “we see that our model comes out on top.”

Green adds, “The CheMeleon foundation model drastically reduces the number of molecules that need to be tested experimentally to accurately predict the properties of new molecules. This is a huge advantage, since synthesizing and testing a large number of molecules is slow and expensive."

Besides the MIT members, the team also included Jochen Sieg, Christian Feldmann, and Miriam Mathea, researchers at BASF SE in Germany. The work was supported by the Machine Learning for Pharmaceutical Discovery and Synthesis consortium, along with fellowship support from U.S. Department of Energy and MathWorks.

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

MIT affiliates elected 2026 American Geophysical Union Fellows

Three MIT affiliates — Anthea Coster of MIT Haystack Observatory, Carolyn Ruppel ’86, SM ’86, PhD ’92 of the MIT Energy Initiative, and César Terrer of the Department of Civil and Environmental Engineering — have been elected as fellows of the American Geophysical Union (AGU) for 2026. 

The AGU also honored eight additional MIT alumni as new fellows: Claudia Benitez-Nelson PhD ’99; Robert Howarth PhD ’79; Douglas J. Jerolmack PhD ’06; Tamara Shapiro Ledley; Galen A. McKinley PhD ’02; Mark David Schmitz PhD ’02; Frederik J. Simons PhD ’02; and Dawn Y. Sumner PhD ’95.

With some 60,000 members in more than 135 countries, the AGU is the world’s largest professional organization for Earth and space scientists and annually elects less than 0.1% of its membership to fellow status. Fellows are chosen from among AGU members who “have made exceptional contributions to Earth and space science through a breakthrough, discovery, or innovation in their field.” AGU Fellows are recognized for their scientific eminence, demonstrated through breakthroughs, discoveries, or innovations that advance the Earth and space sciences. Fellows also act as external experts, advising government agencies and other organizations outside the sciences.

Anthea Coster

Anthea J. Coster is an MIT principal research scientist and Haystack Observatory assistant director emerita.

Coster joined MIT in 1984, originally at MIT Lincoln Laboratory. At Haystack, her career work focuses on atmospheric physics across the ionosphere, magnetosphere, and thermosphere, along with space weather effects and coupling across these different parts of the complex Earth system. She has been responsible for numerous breakthrough contributions in global navigation satellite signal (GNSS) applications to geospace research, including the first GNSS observations of storm-enhanced density across the heavily populated continental United States. She is a recent winner of the Appleton Prize from the International Radio Science Union.

One of the nomination letters for Coster’s AGU Fellowship states, “recognizing that high-level conferences alone were insufficient to provide the necessary training for researchers in developing nations, she co-founded the annual GNSS workshop series (since 2009) in partnership with Boston College and ICTP [the Abdus Salam International Center for Theoretical Physics], training 40–50 participants yearly from developing countries … Through the U.N.-sponsored International Space Weather Initiative (ISWI) and other programs, she has participated in numerous public outreach initiatives, consistently working to make space science accessible to diverse audiences worldwide.” Her nomination also mentioned “[h]er efforts in bringing space physics to South America, Africa, and tribal Alaska ... as does her work with advocating for opportunities for women and underserved populations,” and highlighted her integrity, work excellence, respect for colleagues, and collaboration efforts, as well as her public outreach activities.

Carolyn Ruppel

Carolyn Ruppel, who earned three degrees from the MIT Department of Earth, Atmospheric and Planetary Sciences, is currently the deputy director of science and technology at the MIT Energy Initiative (MITEI). Ruppel, a life member of the AGU, is an expert on gas hydrates, an icelike combination of water and concentrated methane that sequesters an estimated 15% of Earth’s massive methane budget. Before joining MITEI in 2025, she held positions at Woods Hole Oceanographic Institution (1992-93), Georgia Tech (1994-2006), and the U.S. Geological Survey’s (USGS’s) Woods Hole coastal and marine research center (2006-2025). In her current position, Ruppel supports Director William H. Green in leading MITEI’s strategic and technical activities. She coordinates among member companies, MIT researchers, and MITEI’s internal teams to manage its portfolio of energy research, education, and outreach.

Starting in 2010, Ruppel led the renowned USGS Gas Hydrates Project, focusing on the resource potential of these shallowly buried and widely distributed deposits, as well as possible feedbacks among warming climate, gas hydrate breakdown, and release of methane (a greenhouse gas) to the ocean and atmosphere. In 2011, Ruppel wrote the supplementary paper on gas hydrates for the MITEI study on “The Future of Natural Gas.” Ruppel also led the influential study debunking catastrophic scenarios for massive gas hydrate breakdown due to global warming and co-discovered over 700 seafloor methane seeps on the U.S. Atlantic margin. Ruppel’s other scientific contributions have been in the fields of continental rifting, subsea permafrost distribution, hydrate reservoir dynamics, and marine heat flow and acoustics.  

César Terrer

César Terrer is the TianFu Group Career Development Professor and associate professor of civil and environmental engineering at MIT.

Terrer studies the land carbon sink: The roughly 30 percent of annual carbon dioxide emissions that plants and soils absorb each year, and whether that will hold through this century. It is one of the largest uncertainties in climate projection. Earth system models disagree about the fate of the land sink by an amount comparable to decades of emissions, largely because they represent the biology that governs it through assumptions rather than measurements. Terrer’s group replaces those assumptions with data, combining global observations — CO₂ experiments, soil inventories, forest plots and satellites. That work has shown that soil nutrients limit how much extra carbon plants take up as CO₂ rises, about three times less than models had assumed; that plant and soil carbon trade off against one another rather than accumulating together as models assume; and that animals are part of the carbon cycle, with tropical forests that have lost their seed dispersers regrowing less than a quarter as much carbon as intact ones. His group is now testing whether soils, rather than vegetation, account for most of the recent land sink. 

All of Terrer’s first-author papers are cited in the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. NASA selected him to co-lead the modeling and data synthesis working group for PANGEA, the scoping study shaping the agency’s next tropical field campaign, and Apple, Cargill, PepsiCo, Inditex, and other organizations have sought his advice on natural climate solutions.

Alumni honored

The eight additional MIT alumni honorees elected AGU Fellows graduated from the following departments:

Claudia Benitez-Nelson PhD ’99 — Earth, Atmospheric and Planetary Sciences (with WHOI);

Robert Howarth PhD ’79 — Biology (with WHOI);

Douglas J. Jerolmack PhD ’06 — Earth, Atmospheric and Planetary Sciences;

Tamara Shapiro Ledley PhD ’83 — Meteorology;

Galen A. McKinley PhD ’02 — Earth, Atmospheric and Planetary Sciences;

Mark David Schmitz PhD ’02 — Earth, Atmospheric, and Planetary Sciences;

Frederik J. Simons PhD ’02 — Earth, Atmospheric, and Planetary Sciences; and

Dawn Y. Sumner PhD ’95 — Earth, Atmospheric, and Planetary Sciences.

AGU Fellow awards will be presented at the AGU Fall Meeting in San Francisco in December.

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

Scene at MIT: Beautiful, intricate, creative quilts

The MIT Quilt Show, put on as a collaboration between the Crafters’ Circle, MIT Makerspaces, and the MIT Women’s League, and funded by the Council for the Arts at MIT, ran for three days in Lobby 13 Oct. 5-7 and drew 950 visitors.

“We got wonderful feedback throughout the show, and it was tremendously gratifying to see the MIT community turn out en masse to support this art form,” says Jane Halpern, show organizer and communications officer for the Department of Electrical Engineering and Computer Science.

The exhibit featured around 40 quilts and quilt-adjacent works from about 20 members of the MIT community, plus collaborative works such as an astrophysics community quilt, assembled by graduate student Swati Ravi, and an American flag featuring the embroidery work of refugees, assembled by Anh Vu Sawyer MBA ’20 and her husband Philip.

While the quilt show was a one-time event, Halpern encourages MIT community members who are interested in adding more quilts to their lives to join the Crafters’ Circle Slack channel; check for class listings in the Makerspaces; and visit the New England Quilt Museum in nearby Lowell, Massachusetts.

de MIT News https://ift.tt/isPN3m2