Login with your Social Account

A battery-free sensor for underwater exploration

A battery-free sensor for underwater exploration

To investigate the vastly unexplored oceans covering most our planet, researchers aim to build a submerged network of interconnected sensors that send data to the surface — an underwater “internet of things.” But how to supply constant power to scores of sensors designed to stay for long durations in the ocean’s deep?

MIT researchers have an answer: a battery-free underwater communication system that uses near-zero power to transmit sensor data. The system could be used to monitor sea temperatures to study climate change and track marine life over long periods — and even sample waters on distant planets. They are presenting the system at the SIGCOMM conference this week, in a paper that has won the conference’s “best paper” award.

The system makes use of two key phenomena. One, called the “piezoelectric effect,” occurs when vibrations in certain materials generate an electrical charge. The other is “backscatter,” a communication technique commonly used for RFID tags, that transmits data by reflecting modulated wireless signals off a tag and back to a reader.

In the researchers’ system, a transmitter sends acoustic waves through water toward a piezoelectric sensor that has stored data. When the wave hits the sensor, the material vibrates and stores the resulting electrical charge. Then the sensor uses the stored energy to reflect a wave back to a receiver — or it doesn’t reflect one at all. Alternating between reflection in that way corresponds to the bits in the transmitted data: For a reflected wave, the receiver decodes a 1; for no reflected wave, the receiver decodes a 0.

“Once you have a way to transmit 1s and 0s, you can send any information,” says co-author Fadel Adib, an assistant professor in the MIT Media Lab and the Department of Electrical Engineering and Computer Science and founding director of the Signal Kinetics Research Group. “Basically, we can communicate with underwater sensors based solely on the incoming sound signals whose energy we are harvesting.”

The researchers demonstrated their Piezo-Acoustic Backscatter System in an MIT pool, using it to collect water temperature and pressure measurements. The system was able to transmit 3 kilobytes per second of accurate data from two sensors simultaneously at a distance of 10 meters between sensor and receiver.

Applications go beyond our own planet. The system, Adib says, could be used to collect data in the recently discovered subsurface ocean on Saturn’s largest moon, Titan. In June, NASA announced the Dragonfly mission to send a rover in 2026 to explore the moon, sampling water reservoirs and other sites.

“How can you put a sensor under the water on Titan that lasts for long periods of time in a place that’s difficult to get energy?” says Adib, who co-wrote the paper with Media Lab researcher JunSu Jang. “Sensors that communicate without a battery open up possibilities for sensing in extreme environments.”


Preventing deformation

Inspiration for the system hit while Adib was watching “Blue Planet,” a nature documentary series exploring various aspects of sea life. Oceans cover about 72 percent of Earth’s surface. “It occurred to me how little we know of the ocean and how marine animals evolve and procreate,” he says. Internet-of-things (IoT) devices could aid that research, “but underwater you can’t use Wi-Fi or Bluetooth signals … and you don’t want to put batteries all over the ocean, because that raises issues with pollution.”

That led Adib to piezoelectric materials, which have been around and used in microphones and other devices for about 150 years. They produce a small voltage in response to vibrations. But that effect is also reversible: Applying voltage causes the material to deform. If placed underwater, that effect produces a pressure wave that travels through the water. They’re often used to detect sunken vessels, fish, and other underwater objects.

“That reversibility is what allows us to develop a very powerful underwater backscatter communication technology,” Adib says.

Communicating relies on preventing the piezoelectric resonator from naturally deforming in response to strain. At the heart of the system is a submerged node, a circuit board that houses a piezoelectric resonator, an energy-harvesting unit, and a microcontroller. Any type of sensor can be integrated into the node by programming the microcontroller. An acoustic projector (transmitter) and underwater listening device, called a hydrophone (receiver), are placed some distance away.

Say the sensor wants to send a 0 bit. When the transmitter sends its acoustic wave at the node, the piezoelectric resonator absorbs the wave and naturally deforms, and the energy harvester stores a little charge from the resulting vibrations. The receiver then sees no reflected signal and decodes a 0.

However, when the sensor wants to send a 1 bit, the nature changes. When the transmitter sends a wave, the microcontroller uses the stored charge to send a little voltage to the piezoelectric resonator. That voltage reorients the material’s structure in a way that stops it from deforming, and instead reflects the wave. Sensing a reflected wave, the receiver decodes a 1.

Long-term deep-sea sensing

The transmitter and receiver must have power but can be planted on ships or buoys, where batteries are easier to replace, or connected to outlets on land. One transmitter and one receiver can gather information from many sensors covering one area or many areas.

“When you’re tracking a marine animal, for instance, you want to track it over a long range and want to keep the sensor on them for a long period of time. You don’t want to worry about the battery running out,” Adib says. “Or, if you want to track temperature gradients in the ocean, you can get information from sensors covering a number of different places.”

Another interesting application is monitoring brine pools, large areas of brine that sit in pools in ocean basins, and are difficult to monitor long-term. They exist, for instance, on the Antarctic Shelf, where salt settles during the formation of sea ice, and could aid in studying melting ice and marine life interaction with the pools. “We could sense what’s happening down there, without needing to keep hauling sensors up when their batteries die,” Adib says.

Polly Huang, a professor of electrical engineering at Taiwan National University, praised the work for its technical novelty and potential impact on environmental science. “This is a cool idea,” Huang says. “It’s not news one uses piezoelectric crystals to harvest energy … [but is the] first time to see it being used as a radio at the same time [which] is unheard of to the sensor network/system research community. Also interesting and unique is the hardware design and fabrication. The circuit and the design of the encapsulation are both sound and interesting.”

While noting that the system still needs more experimentation, especially in sea water, Huang adds that “this might be the ultimate solution for researchers in marine biography, oceanography, or even meteorology — those in need of long-term, low-human-effort underwater sensing.”

Next, the researchers aim to demonstrate that the system can work at farther distances and communicate with more sensors simultaneously. They’re also hoping to test if the system can transmit sound and low-resolution images.

Physicists design an experiment to pin down the origin of the elements

Physicists design an experiment to pin down the origin of the elements

Nearly all of the oxygen in our universe is forged in the bellies of massive stars like our sun. As these stars contract and burn, they set off thermonuclear reactions within their cores, where nuclei of carbon and helium can collide and fuse in a rare though essential nuclear reaction that generates much of the oxygen in the universe.

The rate of this oxygen-generating reaction has been incredibly tricky to pin down. But if researchers can get a good enough estimate of what’s known as the “radiative capture reaction rate,” they can begin to work out the answers to fundamental questions, such as the ratio of carbon to oxygen in the universe. An accurate rate might also help them determine whether an exploding star will settle into the form of a black hole or a neutron star.

Now physicists at MIT’s Laboratory for Nuclear Science (LNS) have come up with an experimental design that could help to nail down the rate of this oxygen-generating reaction. The approach requires a type of particle accelerator that is still under construction, in several locations around the world. Once up and running, such “multimegawatt” linear accelerators may provide just the right conditions to run the oxgen-generating reaction in reverse, as if turning back the clock of star formation.

The researchers say such an “inverse reaction” should give them an estimate of the reaction rate that actually occurs in stars, with higher accuracy than has previously been achieved.

“The job description of a physicist is to understand the world, and right now, we don’t quite understand where the oxygen in the universe comes from, and, how oxygen and carbon are made,” says Richard Milner, professor of physics at MIT. “If we’re right, this measurement will help us answer some of these important questions in nuclear physics regarding the origin of the elements.”

Milner is a co-author of a paper appearing today in the journal Physical Review C, along with lead author and MIT-LNS postdoc Ivica Friščić and MIT Center for Theoretical Physics Senior Research Scientist T. William Donnelly.

A precipitous drop

The radiative capture reaction rate refers to the reaction between a carbon-12 nucleus and a helium nucleus, also known as an alpha particle, that takes place within a star. When these two nuclei collide, the carbon nucleus effectively “captures” the alpha particle, and in the process, is excited and radiates energy in the form of a photon. What’s left behind is an oxygen-16 nucleus, which ultimately decays to a stable form of oxygen that exists in our atmosphere.

But the chances of this reaction occurring naturally in a star are incredibly slim, due to the fact that both an alpha particle and a carbon-12 nucleus are highly positively charged. If they do come in close contact, they are naturally inclined to repel, in what’s known as a Coulomb’s force. To fuse to form oxygen, the pair would have to collide at sufficiently high energies to overcome Coulomb’s force — a rare occurrence. Such an exceedingly low reaction rate would be impossible to detect at the energy levels that exist within stars.

For the past five decades, scientists have attempted to simulate the radiative capture reaction rate, in small yet powerful particle accelerators. They do so by colliding beams of helium and carbon in hopes of fusing nuclei from both beams to produce oxygen. They have been able to measure such reactions and calculate the associated reaction rates. However, the energies at which such accelerators collide particles are far higher than what occurs in a star, so much so that the current estimates of the oxygen-generating reaction rate are difficult to extrapolate to what actually occurs within stars.

“This reaction is rather well-known at higher energies, but it drops off precipitously as you go down in energy, toward the interesting astrophysical region,” Friščić says.

Time, in reverse

In the new study, the team decided to resurrect a previous notion, to produce the inverse of the oxygen-generating reaction. The aim, essentially, is to start from oxygen gas and split its nucleus into its starting ingredients: an alpha particle and a carbon-12 nucleus. The team reasoned that the probability of the reaction happening in reverse should be greater, and therefore more easily measured, than the same reaction run forward. The inverse reaction should also be possible at energies nearer to the energy range within actual stars.

In order to split oxygen, they would need a high-intensity beam, with a super-high concentration of electrons. (The more electrons that bombard a cloud of oxygen atoms, the more chance there is that one electron among billions will have just the right energy and momentum to collide with and split an oxygen nucleus.)

The idea originated with fellow MIT Research Scientist Genya Tsentalovich, who led a proposed experiment at the MIT-Bates South Hall electron storage ring in 2000.  Although the experiment was never carried out at the Bates accelerator, which ceased operation in 2005, Donnelly and Milner felt the idea merited to be studed in detail. With the initiation of construction of next-generation linear accelerators in Germany and at Cornell University, having the capability to produce electron beams of high enough intensity, or current, to potentially trigger the inverse reaction, and the arrival of Friščić at MIT in 2016, the study got underway.

“The possibility of these new, high-intensity electron machines, with tens of milliamps of current, reawakened our interest in this [inverse reaction] idea,” Milner says.

The team proposed an experiment to produce the inverse reaction by shooting a beam of electrons at a cold, ultradense cloud of oxygen. If an electron successfully collided with and split an oxygen atom, it should scatter away with a certain amount of energy, which physicists have previously predicted. The researchers would isolate the collisions involving electrons within this given energy range, and from these, they would isolate the alpha particles produced in the aftermath.

Alpha particles are produced when O-16 atoms split. The splitting of other oxygen isotopes can also result in alpha particles, but these would scatter away slightly faster — about 10 nanoseconds faster — than alpha particles produced from the splitting of O-16 atoms. So, the team reasoned they would isolate those alpha particles that were slightly slower, with a slightly shorter “time of flight.”

The researchers could then calculate the rate of the inverse reaction, given how often slower alpha particles — and by proxy, the splitting of O-16 atoms — occurred. They then developed a model to relate the inverse reaction to the direct, forward reaction of oxygen production that naturally occurs in stars.

“We’re essentially doing the time-reverse reaction,” Milner says. “If you measure that at the precision we’re talking about, you should be able to directly extract the reaction rate, by factors of  up to 20 beyond what anybody has done in this region.”

Currently, a multimegawatt linear accerator, MESA, is under construction in Germany.  Friščić and Milner are collaborating with physicists there to design the experiment, in hopes that, once up and running, they can put their experiment into action to truly pin down the rate at which stars churn oxygen out into the universe.

“If we’re right, and we make this measurement, it will allow us to answer how much carbon and oxygen is formed in stars, which is the largest uncertainty that we have in our understanding of how stars evolve,” Milner says.

Using Wall Street secrets to reduce the cost of cloud infrastructure

Using Wall Street secrets to reduce the cost of cloud infrastructure

Stock market investors often rely on financial risk theories that help them maximize returns while minimizing financial loss due to market fluctuations. These theories help investors maintain a balanced portfolio to ensure they’ll never lose more money than they’re willing to part with at any given time.

Inspired by those theories, MIT researchers in collaboration with Microsoft have developed a “risk-aware” mathematical model that could improve the performance of cloud-computing networks across the globe. Notably, cloud infrastructure is extremely expensive and consumes a lot of the world’s energy.

Their model takes into account failure probabilities of links between data centers worldwide — akin to predicting the volatility of stocks. Then, it runs an optimization engine to allocate traffic through optimal paths to minimize loss, while maximizing overall usage of the network.

The model could help major cloud-service providers — such as Microsoft, Amazon, and Google — better utilize their infrastructure. The conventional approach is to keep links idle to handle unexpected traffic shifts resulting from link failures, which is a waste of energy, bandwidth, and other resources. The new model, called TeaVar, on the other hand, guarantees that for a target percentage of time — say, 99.9 percent — the network can handle all data traffic, so there is no need to keep any links idle. During that 0.01 percent of time, the model also keeps the data dropped as low as possible.

In experiments based on real-world data, the model supported three times the traffic throughput as traditional traffic-engineering methods, while maintaining the same high level of network availability. A paper describing the model and results will be presented at the ACM SIGCOMM conference this week.

Better network utilization can save service providers millions of dollars, but benefits will “trickle down” to consumers, says co-author Manya Ghobadi, the TIBCO Career Development Assistant Professor in the MIT Department of Electrical Engineering and Computer Science and a researcher at the Computer Science and Artificial Intelligence Laboratory (CSAIL).

“Having greater utilized infrastructure isn’t just good for cloud services — it’s also better for the world,” Ghobadi says. “Companies don’t have to purchase as much infrastructure to sell services to customers. Plus, being able to efficiently utilize datacenter resources can save enormous amounts of energy consumption by the cloud infrastructure. So, there are benefits both for the users and the environment at the same time.”

Joining Ghobadi on the paper are her students Jeremy Bogle and Nikhil Bhatia, both of CSAIL; Ishai Menache and Nikolaj Bjorner of Microsoft Research; and Asaf Valadarsky and Michael Schapira of Hebrew University.

On the money

Cloud service providers use networks of fiber optical cables running underground, connecting data centers in different cities. To route traffic, the providers rely on “traffic engineering” (TE) software that optimally allocates data bandwidth — amount of data that can be transferred at one time — through all network paths.

The goal is to ensure maximum availability to users around the world. But that’s challenging when some links can fail unexpectedly, due to drops in optical signal quality resulting from outages or lines cut during construction, among other factors. To stay robust to failure, providers keep many links at very low utilization, lying in wait to absorb full data loads from downed links.

Thus, it’s a tricky tradeoff between network availability and utilization, which would enable higher data throughputs. And that’s where traditional TE methods fail, the researchers say. They find optimal paths based on various factors, but never quantify the reliability of links. “They don’t say, ‘This link has a higher probability of being up and running, so that means you should be sending more traffic here,” Bogle says. “Most links in a network are operating at low utilization and aren’t sending as much traffic as they could be sending.”

The researchers instead designed a TE model that adapts core mathematics from “conditional value at risk,” a risk-assessment measure that quantifies the average loss of money. With investing in stocks, if you have a one-day 99 percent conditional value at risk of $50, your expected loss of the worst-case 1 percent scenario on that day is $50. But 99 percent of the time, you’ll do much better. That measure is used for investing in the stock market — which is notoriously difficult to predict.

“But the math is actually a better fit for our cloud infrastructure setting,” Ghobadi says. “Mostly, link failures are due to the age of equipment, so the probabilities of failure don’t change much over time. That means our probabilities are more reliable, compared to the stock market.”

Risk-aware model

In networks, data bandwidth shares are analogous to invested “money,” and the network equipment with different probabilities of failure are the “stocks” and their uncertainty of changing values. Using the underlying formulas, the researchers designed a “risk-aware” model that, like its financial counterpart, guarantees data will reach its destination 99.9 percent of time, but keeps traffic loss at minimum during 0.1 percent worst-case failure scenarios. That allows cloud providers to tune the availability-utilization tradeoff.

The researchers statistically mapped three years’ worth of network signal strength from Microsoft’s networks that connects its data centers to a probability distribution of link failures. The input is the network topology in a graph, with source-destination flows of data connected through lines (links) and nodes (cities), with each link assigned a bandwidth.

Failure probabilities were obtained by checking the signal quality of every link every 15 minutes. If the signal quality ever dipped below a receiving threshold, they considered that a link failure. Anything above meant the link was up and running. From that, the model generated an average time that each link was up or down, and calculated a failure probability — or “risk” — for each link at each 15-minute time window. From those data, it was able to predict when risky links would fail at any given window of time.

The researchers tested the model against other TE software on simulated traffic sent through networks from Google, IBM, ATT, and others that spread across the world. The researchers created various failure scenarios based on their probability of occurrence. Then, they sent simulated and real-world data demands through the network and cued their models to start allocating bandwidth.

The researchers’ model kept reliable links working to near full capacity, while steering data clear of riskier links. Over traditional approaches, their model ran three times as much data through the network, while still ensuring all data got to its destination. The code is freely available on GitHub.

Materials provided by Massachusetts Institute of Technology

A new way to deliver drugs with pinpoint targeting

A new way to deliver drugs with pinpoint targeting

Most pharmaceuticals must either be ingested or injected into the body to do their work. Either way, it takes some time for them to reach their intended targets, and they also tend to spread out to other areas of the body. Now, researchers at MIT and elsewhere have developed a system to deliver medical treatments that can be released at precise times, minimally-invasively, and that ultimately could also deliver those drugs to specifically targeted areas such as a specific group of neurons in the brain.

The new approach is based on the use of tiny magnetic particles enclosed within a tiny hollow bubble of lipids (fatty molecules) filled with water, known as a liposome. The drug of choice is encapsulated within these bubbles, and can be released by applying a magnetic field to heat up the particles, allowing the drug to escape from the liposome and into the surrounding tissue.

The findings are reported today in the journal Nature Nanotechnology in a paper by MIT postdoc Siyuan Rao, Associate Professor Polina Anikeeva, and 14 others at MIT, Stanford University, Harvard University, and the Swiss Federal Institute of Technology in Zurich.

“We wanted a system that could deliver a drug with temporal precision, and could eventually target a particular location,” Anikeeva explains. “And if we don’t want it to be invasive, we need to find a non-invasive way to trigger the release.”

Magnetic fields, which can easily penetrate through the body — as demonstrated by detailed internal images produced by magnetic resonance imaging, or MRI — were a natural choice. The hard part was finding materials that could be triggered to heat up by using a very weak magnetic field (about one-hundredth the strength of that used for MRI), in order to prevent damage to the drug or surrounding tissues, Rao says.

Rao came up with the idea of taking magnetic nanoparticles, which had already been shown to be capable of being heated by placing them in a magnetic field, and packing them into these spheres called liposomes. These are like little bubbles of lipids, which naturally form a spherical double layer surrounding a water droplet.

When placed inside a high-frequency but low-strength magnetic field, the nanoparticles heat up, warming the lipids and making them undergo a transition from solid to liquid, which makes the layer more porous — just enough to let some of the drug molecules escape into the surrounding areas. When the magnetic field is switched off, the lipids re-solidify, preventing further releases. Over time, this process can be repeated, thus releasing doses of the enclosed drug at precisely controlled intervals.

The drug carriers were engineered to be stable inside the body at the normal body temperature of 37 degrees Celsius, but able to release their payload of drugs at a temperature of 42 degrees. “So we have a magnetic switch for drug delivery,” and that amount of heat is small enough “so that you don’t cause thermal damage to tissues,” says Anikeeva, who holds appointments in the departments of Materials Science and Engineering and the Brain and Cognitive Sciences.

In principle, this technique could also be used to guide the particles to specific, pinpoint locations in the body, using gradients of magnetic fields to push them along, but that aspect of the work is an ongoing project. For now, the researchers have been injecting the particles directly into the target locations, and using the magnetic fields to control the timing of drug releases. “The technology will allow us to address the spatial aspect,” Anikeeva says, but that has not yet been demonstrated.

This could enable very precise treatments for a wide variety of conditions, she says. “Many brain disorders are characterized by erroneous activity of certain cells. When neurons are too active or not active enough, that manifests as a disorder, such as Parkinson’s, or depression, or epilepsy.” If a medical team wanted to deliver a drug to a specific patch of neurons and at a particular time, such as when an onset of symptoms is detected, without subjecting the rest of the brain to that drug, this system “could give us a very precise way to treat those conditions,” she says.

Rao says that making these nanoparticle-activated liposomes is actually quite a simple process. “We can prepare the liposomes with the particles within minutes in the lab,” she says, and the process should be “very easy to scale up” for manufacturing. And the system is broadly applicable for drug delivery: “we can encapsulate any water-soluble drug,” and with some adaptations, other drugs as well, she says.

One key to developing this system was perfecting and calibrating a way of making liposomes of a highly uniform size and composition. This involves mixing a water base with the fatty acid lipid molecules and magnetic nanoparticles and homogenizing them under precisely controlled conditions. Anikeeva compares it to shaking a bottle of salad dressing to get the oil and vinegar mixed, but controlling the timing, direction and strength of the shaking to ensure a precise mixing.

Anikeeva says that while her team has focused on neurological disorders, as that is their specialty, the drug delivery system is actually quite general and could be applied to almost any part of the body, for example to deliver cancer drugs, or even to deliver painkillers directly to an affected area instead of delivering them systemically and affecting the whole body. “This could deliver it to where it’s needed, and not deliver it continuously,” but only as needed.

Because the magnetic particles themselves are similar to those already in widespread use as contrast agents for MRI scans, the regulatory approval process for their use may be simplified, as their biological compatibility has largely been proven.

The team included researchers in MIT’s departments of Materials Science and Engineering and Brain and Cognitive Sciences, as well as the McGovern Institute for Brain Research, the Simons Center for Social Brain, and the Research Laboratory of Electronics; the Harvard University Department of Chemistry and Chemical Biology and the John A. Paulsen School of Engineering and Applied Sciences; Stanford University; and the Swiss Federal Institute of Technology in Zurich. The work was supported by the Simons Postdoctoral Fellowship, the U.S. Defense Advanced Research Projects Agency, the Bose Research Grant, and the National Institutes of Health.

Materials provided by Massachusetts Institute of Technology

New type of electrolyte could enhance supercapacitor performance

New type of electrolyte could enhance supercapacitor performance

Supercapacitors, electrical devices that store and release energy, need a layer of electrolyte — an electrically conductive material that can be solid, liquid, or somewhere in between. Now, researchers at MIT and several other institutions have developed a novel class of liquids that may open up new possibilities for improving the efficiency and stability of such devices while reducing their flammability.

“This proof-of-concept work represents a new paradigm for electrochemical energy storage,” the researchers say in their paper describing the finding, which appears today in the journal Nature Materials.

For decades, researchers have been aware of a class of materials known as ionic liquids — essentially, liquid salts — but this team has now added to these liquids a compound that is similar to a surfactant, like those used to disperse oil spills. With the addition of this material, the ionic liquids “have very new and strange properties,” including becoming highly viscous, says MIT postdoc Xianwen Mao PhD ’14, the lead author of the paper.

“It’s hard to imagine that this viscous liquid could be used for energy storage,” Mao says, “but what we find is that once we raise the temperature, it can store more energy, and more than many other electrolytes.”

That’s not entirely surprising, he says, since with other ionic liquids, as temperature increases, “the viscosity decreases and the energy-storage capacity increases.” But in this case, although the viscosity stays higher than that of other known electrolytes, the capacity increases very quickly with increasing temperature. That ends up giving the material an overall energy density — a measure of its ability to store electricity in a given volume — that exceeds those of many conventional electrolytes, and with greater stability and safety.

The key to its effectiveness is the way the molecules within the liquid automatically line themselves up, ending up in a layered configuration on the metal electrode surface. The molecules, which have a kind of tail on one end, line up with the heads facing outward toward the electrode or away from it, and the tails all cluster in the middle, forming a kind of sandwich. This is described as a self-assembled nanostructure.

“The reason why it’s behaving so differently” from conventional electrolytes is because of the way the molecules intrinsically assemble themselves into an ordered, layered structure where they come in contact with another material, such as the electrode inside a supercapacitor, says T. Alan Hatton, a professor of chemical engineering at MIT and the paper’s senior author. “It forms a very interesting, sandwich-like, double-layer structure.”

This highly ordered structure helps to prevent a phenomenon called “overscreening” that can occur with other ionic liquids, in which the first layer of ions (electrically charged atoms or molecules) that collect on an electrode surface contains more ions than there are corresponding charges on the surface. This can cause a more scattered distribution of ions, or a thicker ion multilayer, and thus a loss of efficiency in energy storage; “whereas with our case, because of the way everything is structured, charges are concentrated within the surface layer,” Hatton says.

The new class of materials, which the researchers call SAILs, for surface-active ionic liquids, could have a variety of applications for high-temperature energy storage, for example for use in hot environments such as in oil drilling or in chemical plants, according to Mao. “Our electrolyte is very safe at high temperatures, and even performs better,” he says. In contrast, some electrolytes used in lithium-ion batteries are quite flammable.

The material could help to improve performance of supercapacitors, Mao says. Such devices can be used to store electrical charge and are sometimes used to supplement battery systems in electric vehicles to provide an extra boost of power. Using the new material instead of a conventional electrolyte in a supercapacitor could increase its energy density by a factor of four or five, Mao says. Using the new electrolyte, future supercapacitors may even be able to store more energy than batteries, he says, potentially even replacing batteries in applications such as electric vehicles, personal electronics, or grid-level energy storage facilities.

The material could also be useful for a variety of emerging separation processes, Mao says. “A lot of newly developed separation processes require electrical control,” in various chemical processing and refining applications and in carbon dioxide capture, for example, as well as resource recovery from waste streams. These ionic liquids, being highly conductive, could be well-suited to many such applications, he says.

The material they initially developed is just an example of a variety of possible SAIL compounds. “The possibilities are almost unlimited,” Mao says. The team will continue to work on different variations and on optimizing its parameters for particular uses. “It might take a few months or years,” he says, “but working on a new class of materials is very exciting to do. There are many possibilities for further optimization.”

The research team included Paul Brown, Yinying Ren, Agilio Padua, and Margarida Costa Gomes at MIT; Ctirad Cervinka at École Normale Supérieure de Lyon, in France; Gavin Hazell and Julian Eastoe at the University of Bristol, in the U.K.; Hua Li and Rob Atkin at the University of Western Australia; and Isabelle Grillo at the Institut Max-von-Laue-Paul-Langevin in Grenoble, France. The researchers dedicate their paper to the memory of Grillo, who recently passed away.

“It is a very exciting result that surface-active ionic liquids (SAILs) with amphiphilic structures can self-assemble on electrode surfaces and enhance charge storage performance at electrified surfaces,” says Yi Cui, a professor of materials science and engineering at Stanford University, who was not associated with this research. “The authors have studied and understood the mechanism. The work here might have a great impact on the design of high energy density supercapacitors, and could also help improve battery performance,” he says.

Nicholas Abbott, the Tisch University Professor at Cornell University, who also was not involved in this work, says “The paper describes a very clever advance in interfacial charge storage, elegantly demonstrating how knowledge of molecular self-assembly at interfaces can be leveraged to address a contemporary technological challenge.”

Materials provided by Massachusetts Institute of Technology

Tissue model reveals role of blood-brain barrier in Alzheimer’s

Tissue model reveals role of blood-brain barrier in Alzheimer’s

Beta-amyloid plaques, the protein aggregates that form in the brains of Alzheimer’s patients, disrupt many brain functions and can kill neurons. They can also damage the blood-brain barrier — the normally tight border that prevents harmful molecules in the bloodstream from entering the brain.

MIT engineers have now developed a tissue model that mimics beta-amyloid’s effects on the blood-brain barrier, and used it to show that this damage can lead molecules such as thrombin, a clotting factor normally found in the bloodstream, to enter the brain and cause additional damage to Alzheimer’s neurons.

“We were able to show clearly in this model that the amyloid-beta secreted by Alzheimer’s disease cells can actually impair barrier function, and once that is impaired, factors are secreted into the brain tissue that can have adverse effects on neuron health,” says Roger Kamm, the Cecil and Ida Green Distinguished Professor of Mechanical and Biological Engineering at MIT.

The researchers also used the tissue model to show that a drug that restores the blood-brain barrier can slow down the cell death seen in Alzheimer’s neurons.

Kamm and Rudolph Tanzi, a professor of neurology at Harvard Medical School and Massachusetts General Hospital, are the senior authors of the study, which appears in the August 12 issue of the journal Advanced Science. MIT postdoc Yoojin Shin is the paper’s lead author.

Barrier breakdown

The blood vessel cells that make up the blood-brain barrier have many specialized proteins that help them to form tight junctions — cellular structures that act as a strong seal between cells.

Alzheimer’s patients often experience damage to brain blood vessels caused by beta-amyloid proteins, an effect known as cerebral amyloid angiopathy (CAA). It is believed that this damage allows harmful molecules to get into the brain more easily. Kamm decided to study this phenomenon, and its role in Alzheimer’s, by modeling brain and blood vessel tissue on a microfluidic chip.

“What we were trying to do from the start was generate a model that we could use to understand the interactions between Alzheimer’s disease neurons and the brain vasculature,” Kamm says. “Given the fact that there’s been so little success in developing therapeutics that are effective against Alzheimer’s, there has been increased attention paid to CAA over the last couple of years.”

His lab began working on this project several years ago, along with researchers at MGH who had engineered neurons to produce large amounts of beta-amyloid proteins, just like the brain cells of Alzheimer’s patients.

Led by Shin, the researchers devised a way to grow these cells in a microfluidic channel, where they produce and secrete beta-amyloid protein. On the same chip, in a parallel channel, the researchers grew brain endothelial cells, which are the cells that form the blood-brain barrier. An empty channel separated the two channels while each tissue type developed.

After 10 days of cell growth, the researchers added collagen to the central channel separating the two tissue types, which allowed molecules to diffuse from one channel to the other. They found that within three to six days, beta-amyloid proteins secreted by the neurons began to accumulate in the endothelial tissue, which led the cells to become leakier. These cells also showed a decline in proteins that form tight junctions, and an increase in enzymes that break down the extracellular matrix that normally surrounds and supports blood vessels.

As a result of this breakdown in the blood-brain barrier, thrombin was able to pass from blood flowing through the leaky vessels into the Alzheimer’s neurons. Excessive levels of thrombin can harm neurons and lead to cell death.

“We were able to demonstrate this bidirectional signaling between cell types and really solidify things that had been seen previously in animal experiments, but reproduce them in a model system that we can control with much more detail and better fidelity,” Kamm says.

Plugging the leaks

The researchers then decided to test two drugs that have previously been shown to solidify the blood-brain barrier in simpler models of endothelial tissue. Both of these drugs are FDA-approved to treat other conditions. The researchers found that one of these drugs, etodolac, worked very well, while the other, beclomethasone, had little effect on leakiness in their tissue model.

In tissue treated with etodolac, the blood-brain barrier became tighter, and neurons’ survival rates improved. The MIT and MGH team is now working with a drug discovery consortium to look for other drugs that might be able to restore the blood-brain barrier in Alzheimer’s patients.

“We’re starting to use this platform to screen for drugs that have come out of very simple single cell screens that we now need to validate in a more complex system,” Kamm says. “This approach could offer a new potential form of Alzheimer’s treatment, especially given the fact that so few treatments have been demonstrated to be effective.”

Materials provided by Massachusetts Institute of Technology

Guided by AI, robotic platform automates molecule manufacture

Guided by AI, robotic platform automates molecule manufacture

Guided by artificial intelligence and powered by a robotic platform, a system developed by MIT researchers moves a step closer to automating the production of small molecules that could be used in medicine, solar energy, and polymer chemistry.

The system, described in the August 8 issue of Science, could free up bench chemists from a variety of routine and time-consuming tasks, and may suggest possibilities for how to make new molecular compounds, according to the study co-leaders Klavs F. Jensen, the Warren K. Lewis Professor of Chemical Engineering, and Timothy F. Jamison, the Robert R. Taylor Professor of Chemistry and associate provost at MIT.

The technology “has the promise to help people cut out all the tedious parts of molecule building,” including looking up potential reaction pathways and building the components of a molecular assembly line each time a new molecule is produced, says Jensen.

“And as a chemist, it may give you inspirations for new reactions that you hadn’t thought about before,” he adds.

Other MIT authors on the Science paper include Connor W. Coley, Dale A. Thomas III, Justin A. M. Lummiss, Jonathan N. Jaworski, Christopher P. Breen, Victor Schultz, Travis Hart, Joshua S. Fishman, Luke Rogers, Hanyu Gao, Robert W. Hicklin, Pieter P. Plehiers, Joshua Byington, John S. Piotti, William H. Green, and A. John Hart.

From inspiration to recipe to finished product

The new system combines three main steps. First, software guided by artificial intelligence suggests a route for synthesizing a molecule, then expert chemists review this route and refine it into a chemical “recipe,” and finally the recipe is sent to a robotic platform that automatically assembles the hardware and performs the reactions that build the molecule.

Coley and his colleagues have been working for more than three years to develop the open-source software suite that suggests and prioritizes possible synthesis routes. At the heart of the software are several neural network models, which the researchers trained on millions of previously published chemical reactions drawn from the Reaxys and U.S. Patent and Trademark Office databases. The software uses these data to identify the reaction transformations and conditions that it believes will be suitable for building a new compound.

“It helps makes high-level decisions about what kinds of intermediates and starting materials to use, and then slightly more detailed analyses about what conditions you might want to use and if those reactions are likely to be successful,” says Coley.

“One of the primary motivations behind the design of the software is that it doesn’t just give you suggestions for molecules we know about or reactions we know about,” he notes. “It can generalize to new molecules that have never been made.”

Chemists then review the suggested synthesis routes produced by the software to build a more complete recipe for the target molecule. The chemists sometimes need to perform lab experiments or tinker with reagent concentrations and reaction temperatures, among other changes.

“They take some of the inspiration from the AI and convert that into an executable recipe file, largely because the chemical literature at present does not have enough information to move directly from inspiration to execution on an automated system,” Jamison says.

The final recipe is then loaded on to a platform where a robotic arm assembles modular reactors, separators, and other processing units into a continuous flow path, connecting pumps and lines that bring in the molecular ingredients.

“You load the recipe — that’s what controls the robotic platform — you load the reagents on, and press go, and that allows you to generate the molecule of interest,” says Thomas. “And then when it’s completed, it flushes the system and you can load the next set of reagents and recipe, and allow it to run.”

Unlike the continuous flow system the researchers presented last year, which had to be manually configured after each synthesis, the new system is entirely configured by the robotic platform.

“This gives us the ability to sequence one molecule after another, as well as generate a library of molecules on the system, autonomously,” says Jensen.

The design for the platform, which is about two cubic meters in size — slightly smaller than a standard chemical fume hood — resembles a telephone switchboard and operator system that moves connections between the modules on the platform.

“The robotic arm is what allowed us to manipulate the fluidic paths, which reduced the number of process modules and fluidic complexity of the system, and by reducing the fluidic complexity we can increase the molecular complexity,” says Thomas. “That allowed us to add additional reaction steps and expand the set of reactions that could be completed on the system within a relatively small footprint.”

Toward full automation

The researchers tested the full system by creating 15 different medicinal small molecules of different synthesis complexity, with processes taking anywhere between two hours for the simplest creations to about 68 hours for manufacturing multiple compounds.

The team synthesized a variety of compounds: aspirin and the antibiotic secnidazole in back-to-back processes; the painkiller lidocaine and the antianxiety drug diazepam in back-to-back processes using a common feedstock of reagents; the blood thinner warfarin and the Parkinson’s disease drug safinamide, to show how the software could design compounds with similar molecular components but differing 3-D structures; and a family of five ACE inhibitor drugs and a family of four nonsteroidal anti-inflammatory drugs.

“I’m particularly proud of the diversity of the chemistry and the kinds of different chemical reactions,” says Jamison, who said the system handled about 30 different reactions compared to about 12 different reactions in the previous continuous flow system.

“We are really trying to close the gap between idea generation from these programs and what it takes to actually run a synthesis,” says Coley. “We hope that next-generation systems will increase further the fraction of time and effort that scientists can focus their efforts on creativity and design.”

Materials provided by Massachusetts Institute of Technology

Optimus Ride’s autonomous system makes self-driving vehicles a reality

Optimus Ride’s autonomous system makes self-driving vehicles a reality

Some of the biggest companies in the world are spending billions in the race to develop self-driving vehicles that can go anywhere. Meanwhile, Optimus Ride, a startup out of MIT, is already helping people get around by taking a different approach.

The company’s autonomous vehicles only drive in areas it comprehensibly maps, or geofences. Self-driving vehicles can safely move through these areas at about 25 miles per hour with today’s technology.

“It’s important to realize there are multiple approaches, and multiple markets, to self-driving,” says Optimus Ride CEO Ryan Chin MA ’00, SM ’04, PhD ’12. “There’s no monolithic George Jetson kind of self-driving vehicle. You have robot trucks, you have self-driving taxis, self-driving pizza delivery machines, and each of these will have different time frames of technological development and different markets.”

By partnering with developers, the Optimus team is currently focused on deploying its vehicles in communities with residential and commercial buildings, retirement communities, corporate and university campuses, airports, resorts, and smart cities. The founders estimate the combined value of transportation services in those markets to be over $600 billion.

“We believe this is an important, huge business, but we also believe this is the first addressable market in the sense that we believe the first autonomous vehicles that will generate profits and make business sense will appear in these environments, because you can build the tech much more quickly,” says Chin, who co-founded the company with Albert Huang SM ’05, PhD ’10, Jenny Larios Berlin MCP ’14, MBA ’15, Ramiro Almeida, and Class of 1948 Career Development Professor of Aeronautics and Astronautics Sertac Karaman.

Optimus Ride currently runs fleets of self-driving vehicles in the Seaport area of Boston, in a mixed-use development in South Weymouth, Massachusetts, and, as of this week, in the Brooklyn Navy Yard, a 300-acre industrial park that now hosts the first self-driving vehicle program in the state.

Later this year, the company will also deploy its autonomous vehicles in a private community of Fairfield, California, and in a mixed-use development in Reston, Virginia.

The early progress — and the valuable data that come with it — is the result of the company taking a holistic view of transportation. That perspective can be traced back to the founders’ diverse areas of focus at MIT.

A multidisciplinary team

Optimus Ride’s founders have worked across a wide array of departments, labs, and centers across MIT. The technical validation for the company began when Karaman participated in the Defense Advanced Research Projects Agency’s (DARPA) Urban Challenge with a team including Huang in 2007. Both researchers had also worked in the Computer Science and Artificial Intelligence Laboratory together.

For the event, DARPA challenged 89 teams with creating a fully autonomous vehicle that could traverse a 60 mile course in under six hours. The vehicle from MIT was one of only six to complete the journey.

Chin, who led a Media Lab project that developed a retractable electric vehicle in the Smart Cities group, met Karaman when both were PhD candidates in 2012. Almeida began working in the Media Lab as a visiting scholar a year later.

As members of the group combined their expertise on both self-driving technology and the way people move around communities, they realized they needed help developing business models around their unique approach to improving transportation. Jenny Larios Berlin was introduced to the founders in 2015 after earning joint degrees from the Department of Urban Studies and Planning and the Sloan School of Management. The team started Optimus Ride in August that year.

“The company is really a melting pot of ideas from all of these schools and departments,” Karaman says. “When we met each other, there was the technology angle, but we also realized there’s an important business angle, and there’s also an interesting urban planning/media arts and sciences angle around thinking of the system as a whole. So when we formed the company we thought, not just how can we build fully autonomous vehicles, but also how can we make transportation in general more affordable, sustainable, equitable, accessible, and so on.”

Karaman says the company’s approach could only have originated in a highly collaborative environment like MIT, and believes it gives the company a big advantage in the self-driving sector.

“I knew how to build autonomous systems, but in interacting with Ryan and Ramiro and Jenny, I really got a better understanding of what the systems would look like, what the smart cities that utilize the systems would look like, what some of the business models would look like,” Karaman says. “That has a feedback on the technology. It allows you to build the right kind of technology very efficiently in order to go to these markets.”

Optimus Ride’s self-driving vehicles can travel on many public roads. Courtesy of Optimus Ride

First mover advantage

Optimus Ride’s vehicles have a suite of cameras, lasers, and sensors similar to what other companies use to help autonomous vehicles navigate their environments. But Karaman says the company’s key technical differentiators are its machine vision system, which rapidly identifies objects, and its ability to fuse all those data sources together to make predictions, such as where an object is going and when it will get there.

Optimus Ride’s vehicles feature a range of cameras and sensors to help them navigate their environment. Courtesy of Optimus Ride

The strictly defined areas where the vehicles drive help them learn what Karaman calls the “culture of driving” on different roads. Human drivers might subconsciously take a little longer at certain intersections. Commuters might drive much faster than the speed limit. Those and other location-specific details, like the turn radius of the Silver Line bus in the Seaport, are learned by the system through experience.

“A lot of the well-funded autonomous driving projects out there try to capture everything at the same time and tackle every problem,” Karaman says. “But we operate the vehicle in places where it can learn very rapidly. If you go around, say, 10,000 miles in a small community, you end up seeing a certain intersection a hundred or a thousand times, so you learn the culture of driving through that intersection. But if you go 10,000 miles around the country, you’ll only see places once.”

Safety drivers are still required to be behind the wheels of autonomous vehicles in the states Optimus Ride operates in, but the founders hope to soon be monitoring fleets with fewer people in a manner similar to an air traffic controller.

For now, though, they’re focused on scaling their current model. The contract in Reston, Virginia is part of a strategic partnership with one of the largest real estate managers in the world, Brookfield Properties. Chin says Brookfield owns over 100 locations where Optimus Ride could deploy its system, and the company is aiming to be operating 10 or more fleets by the end of 2020.

“Collectively, [the founders] probably have around three decades of experience in building self-driving vehicles, electric vehicles, shared vehicles, mobility transportation, on demand systems, and in looking at how you integrate new transportation systems into cities,” Chin says. “So that’s been the idea of the company: to marry together technical expertise with the right kind of policymaking, the right kind of business models, and to bring autonomy to the world as fast as possible.”

Materials provided by Massachusetts Institute of Technology
Study furthers radically new view of gene control

Study furthers radically new view of gene control

In recent years, MIT scientists have developed a new model for how key genes are controlled that suggests the cellular machinery that transcribes DNA into RNA forms specialized droplets called condensates. These droplets occur only at certain sites on the genome, helping to determine which genes are expressed in different types of cells.

In a new study that supports that model, researchers at MIT and the Whitehead Institute for Biomedical Research have discovered physical interactions between proteins and with DNA that help explain why these droplets, which stimulate the transcription of nearby genes, tend to cluster along specific stretches of DNA known as super-enhancers. These enhancer regions do not encode proteins but instead, regulate other genes.

“This study provides a fundamentally important new approach to deciphering how the ‘dark matter’ in our genome functions in gene control,” says Richard Young, an MIT professor of biology and member of the Whitehead Institute.

Young is one of the senior authors of the paper, along with Phillip Sharp, an MIT Institute Professor and member of MIT’s Koch Institute for Integrative Cancer Research; and Arup K. Chakraborty, the Robert T. Haslam Professor in Chemical Engineering, a professor of physics and chemistry, and a member of MIT’s Institute for Medical Engineering and Science and the Ragon Institute of MGH, MIT, and Harvard.

Graduate student Krishna Shrinivas and postdoc Benjamin Sabari are the lead authors of the paper, which appears in Molecular Cell on Aug. 8.

“A biochemical factory”

Every cell in an organism has an identical genome, but cells such as neurons or heart cells express different subsets of those genes, allowing them to carry out their specialized functions. Previous research has shown that many of these genes are located near super enhancers, which bind to proteins called transcription factors that stimulate the copying of nearby genes into RNA.

About three years ago, Sharp, Young, and Chakraborty joined forces to try to model the interactions that occur at enhancers. In a 2017 Cell paper, based on computational studies, they hypothesized that in these regions, transcription factors form droplets called phase-separated condensates. Similar to droplets of oil suspended in salad dressing, these condensates are collections of molecules that form distinct cellular compartments but have no membrane separating them from the rest of the cell.

In a 2018 Science paper, the researchers showed that these dynamic droplets do form at super enhancer locations. Made of clusters of transcription factors and other molecules, these droplets attract enzymes such as RNA polymerases that are needed to copy DNA into messenger RNA, keeping gene transcription active at specific sites.

“We had demonstrated that the transcription machinery forms liquid-like droplets at certain regulatory regions on our genome, however we didn’t fully understand how or why these dewdrops of biological molecules only seemed to condense around specific points on our genome,” Shrinivas says.

As one possible explanation for that site specificity, the research team hypothesized that weak interactions between intrinsically disordered regions of transcription factors and other transcriptional molecules, along with specific interactions between transcription factors and particular DNA elements, might determine whether a condensate forms at a particular stretch of DNA. Biologists have traditionally focused on “lock-and-key” style interactions between rigidly structured protein segments to explain most cellular processes, but more recent evidence suggests that weak interactions between floppy protein regions also play an important role in cell activities.

In this study, computational modeling and experimentation revealed that the cumulative force of these weak interactions conspire together with transcription factor-DNA interactions to determine whether a condensate of transcription factors will form at a particular site on the genome. Different cell types produce different transcription factors, which bind to different enhancers. When many transcription factors cluster around the same enhancers, weak interactions between the proteins are more likely to occur. Once a critical threshold concentration is reached, condensates form.

“Creating these local high concentrations within the crowded environment of the cell enables the right material to be in the right place at the right time to carry out the multiple steps required to activate a gene,” Sabari says. “Our current study begins to tease apart how certain regions of the genome are capable of pulling off this trick.”

These droplets form on a timescale of seconds to minutes, and they blink in and out of existence depending on a cell’s needs.

“It’s an on-demand biochemical factory that cells can form and dissolve, as and when they need it,” Chakraborty says. “When certain signals happen at the right locus on a gene, the condensates form, which concentrates all of the transcription molecules. Transcription happens, and when the cells are done with that task, they get rid of them.”

A new view

Weak cooperative interactions between proteins may also play an important role in evolution, the researchers proposed in a 2018 Proceedings of the National Academy of Sciences paper. The sequences of intrinsically disordered regions of transcription factors need to change only a little to evolve new types of specific functionality. In contrast, evolving new specific functions via “lock-and-key” interactions requires much more significant changes.

“If you think about how biological systems have evolved, they have been able to respond to different conditions without creating new genes. We don’t have any more genes that a fruit fly, yet we’re much more complex in many of our functions,” Sharp says. “The incremental expanding and contracting of these intrinsically disordered domains could explain a large part of how that evolution happens.”

Similar condensates appear to play a variety of other roles in biological systems, offering a new way to look at how the interior of a cell is organized. Instead of floating through the cytoplasm and randomly bumping into other molecules, proteins involved in processes such as relaying molecular signals may transiently form droplets that help them interact with the right partners.

“This is a very exciting turn in the field of cell biology,” Sharp says. “It is a whole new way of looking at biological systems that is richer and more meaningful.”

Some of the MIT researchers, led by Young, have helped form a company called Dewpoint Therapeutics to develop potential treatments for a wide variety of diseases by exploiting cellular condensates. There is emerging evidence that cancer cells use condensates to control sets of genes that promote cancer, and condensates have also been linked to neurodegenerative disorders such as amyotrophic lateral sclerosis (ALS) and Huntington’s disease.

Materials provided by Massachusetts Institute of Technology

Automating artificial intelligence for medical decision-making

Automating artificial intelligence for medical decision-making

MIT computer scientists are hoping to accelerate the use of artificial intelligence to improve medical decision-making, by automating a key step that’s usually done by hand — and that’s becoming more laborious as certain datasets grow ever-larger.

The field of predictive analytics holds increasing promise for helping clinicians diagnose and treat patients. Machine-learning models can be trained to find patterns in patient data to aid in sepsis care, design safer chemotherapy regimens, and predict a patient’s risk of having breast cancer or dying in the ICU, to name just a few examples.

Typically, training datasets consist of many sick and healthy subjects, but with relatively little data for each subject. Experts must then find just those aspects — or “features” — in the datasets that will be important for making predictions.

This “feature engineering” can be a laborious and expensive process. But it’s becoming even more challenging with the rise of wearable sensors, because researchers can more easily monitor patients’ biometrics over long periods, tracking sleeping patterns, gait, and voice activity, for example. After only a week’s worth of monitoring, experts could have several billion data samples for each subject.

In a paper being presented at the Machine Learning for Healthcare conference this week, MIT researchers demonstrate a model that automatically learns features predictive of vocal cord disorders. The features come from a dataset of about 100 subjects, each with about a week’s worth of voice-monitoring data and several billion samples — in other words, a small number of subjects and a large amount of data per subject. The dataset contain signals captured from a little accelerometer sensor mounted on subjects’ necks.

In experiments, the model used features automatically extracted from these data to classify, with high accuracy, patients with and without vocal cord nodules. These are lesions that develop in the larynx, often because of patterns of voice misuse such as belting out songs or yelling. Importantly, the model accomplished this task without a large set of hand-labeled data.

“It’s becoming increasing easy to collect long time-series datasets. But you have physicians that need to apply their knowledge to labeling the dataset,” says lead author Jose Javier Gonzalez Ortiz, a PhD student in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). “We want to remove that manual part for the experts and offload all feature engineering to a machine-learning model.”

The model can be adapted to learn patterns of any disease or condition. But the ability to detect the daily voice-usage patterns associated with vocal cord nodules is an important step in developing improved methods to prevent, diagnose, and treat the disorder, the researchers say. That could include designing new ways to identify and alert people to potentially damaging vocal behaviors.

Joining Gonzalez Ortiz on the paper is John Guttag, the Dugald C. Jackson Professor of Computer Science and Electrical Engineering and head of CSAIL’s Data Driven Inference Group; Robert Hillman, Jarrad Van Stan, and Daryush Mehta, all of Massachusetts General Hospital’s Center for Laryngeal Surgery and Voice Rehabilitation; and Marzyeh Ghassemi, an assistant professor of computer science and medicine at the University of Toronto.

Forced feature-learning

For years, the MIT researchers have worked with the Center for Laryngeal Surgery and Voice Rehabilitation to develop and analyze data from a sensor to track subject voice usage during all waking hours. The sensor is an accelerometer with a node that sticks to the neck and is connected to a smartphone. As the person talks, the smartphone gathers data from the displacements in the accelerometer.

In their work, the researchers collected a week’s worth of this data — called “time-series” data — from 104 subjects, half of whom were diagnosed with vocal cord nodules. For each patient, there was also a matching control, meaning a healthy subject of similar age, sex, occupation, and other factors.

Traditionally, experts would need to manually identify features that may be useful for a model to detect various diseases or conditions. That helps prevent a common machine-learning problem in health care: overfitting. That’s when, in training, a model “memorizes” subject data instead of learning just the clinically relevant features. In testing, those models often fail to discern similar patterns in previously unseen subjects.

“Instead of learning features that are clinically significant, a model sees patterns and says, ‘This is Sarah, and I know Sarah is healthy, and this is Peter, who has a vocal cord nodule.’ So, it’s just memorizing patterns of subjects. Then, when it sees data from Andrew, which has a new vocal usage pattern, it can’t figure out if those patterns match a classification,” Gonzalez Ortiz says.

The main challenge, then, was preventing overfitting while automating manual feature engineering. To that end, the researchers forced the model to learn features without subject information. For their task, that meant capturing all moments when subjects speak and the intensity of their voices.

As their model crawls through a subject’s data, it’s programmed to locate voicing segments, which comprise only roughly 10 percent of the data. For each of these voicing windows, the model computes a spectrogram, a visual representation of the spectrum of frequencies varying over time, which is often used for speech processing tasks. The spectrograms are then stored as large matrices of thousands of values.

But those matrices are huge and difficult to process. So, an autoencoder — a neural network optimized to generate efficient data encodings from large amounts of data — first compresses the spectrogram into an encoding of 30 values. It then decompresses that encoding into a separate spectrogram.

Basically, the model must ensure that the decompressed spectrogram closely resembles the original spectrogram input. In doing so, it’s forced to learn the compressed representation of every spectrogram segment input over each subject’s entire time-series data. The compressed representations are the features that help train machine-learning models to make predictions.

Mapping normal and abnormal features

In training, the model learns to map those features to “patients” or “controls.” Patients will have more voicing patterns than will controls. In testing on previously unseen subjects, the model similarly condenses all spectrogram segments into a reduced set of features. Then, it’s majority rules: If the subject has mostly abnormal voicing segments, they’re classified as patients; if they have mostly normal ones, they’re classified as controls.

In experiments, the model performed as accurately as state-of-the-art models that require manual feature engineering. Importantly, the researchers’ model performed accurately in both training and testing, indicating it’s learning clinically relevant patterns from the data, not subject-specific information.

Next, the researchers want to monitor how various treatments — such as surgery and vocal therapy — impact vocal behavior. If patients’ behaviors move form abnormal to normal over time, they’re most likely improving. They also hope to use a similar technique on electrocardiogram data, which is used to track muscular functions of the heart.

Materials provided by Massachusetts Institute of Technology