After multiple delays of the REAL ID Act of 2005 and its updated counterpart, the REAL ID Modernization Act, in the United States, the May 7th deadline of REAL ID enforcement has finally arrived. Does this move our security forward in the skies? The last 20 years says we got along fine without it. There were and are issues along the way that REAL ID does impose on everyday people, such as potential additional costs and rigid documentation, even if you already have a state issued ID. While TSA states this is not a national ID or a federal database, but a set of minimum standards required for federal use, we are still watchful of the mechanisms that have pivoted to potential privacy issues with the expansion of digital IDs.

But you don’t need a REAL ID just to fly domestically. There are alternatives.

The most common alternatives are passports or passport cards. You can use either instead of a REAL ID, which might save you an immediate trip to the DMV. And the additional money for a passport at least provides you the extra benefit of international travel.

Passports and passport cards are not the only alternatives to REAL ID. Additional documentation is also accepted as well: (this list is subject to change by the TSA):

• REAL ID-compliant driver's licenses or other state photo identity cards issued by Department of Motor Vehicles (or equivalent and this excludes a temporary driver’s license)
• State-issued Enhanced Driver's License (EDL) or Enhanced ID (EID)
• U.S. passport
• U.S. passport card
• DHS trusted traveler cards (Global Entry, NEXUS, SENTRI, FAST)
• U.S. Department of Defense ID, including IDs issued to dependents
• Permanent resident card
• Border crossing card
• An acceptable photo ID issued by a federally recognized Tribal Nation/Indian Tribe, including Enhanced Tribal Cards (ETCs)
• HSPD-12 PIV card
• Foreign government-issued passport
• Canadian provincial driver's license or Indian and Northern Affairs Canada card
• Transportation Worker Identification Credential (TWIC)
• U.S. Citizenship and Immigration Services Employment Authorization Card (I-766)
• U.S. Merchant Mariner Credential
• Veteran Health Identification Card (VHIC)

Foreign government-issued passports are on this list. However, using a foreign-government issued passport may increase your chances of closer scrutiny at the security gate. REAL ID and other federally accepted documents are supposed to be about verifying your identity, not about your citizenship status. Realistically, interactions with secondary screening and law enforcement are not out of the realm of possibility for non-citizens. The power dynamics of the border has now been brought to flying domestically thanks to REAL ID. The privileges of who can and can’t fly are more sensitive now.

REAL ID and other federally accepted documents are supposed to be about verifying your identity, not about your citizenship status

Mobile Driver’s Licenses (mDLs)

Many states have rolled out the option for a Mobile Driver's License, which acts as a form of your state-issued ID on your phone and is supposed to come with an exception for REAL ID compliance. This is something we asked for since mDLs appear to satisfy their fears of forgery and cloning. But the catch is that states had to apply for this waiver:

“The final rule, effective November 25, 2024, allows states to apply to TSA for a temporary waiver of certain REAL ID requirements written in the REAL ID regulations.”

TSA stated they would publish the list of states with this waiver. But we do not see it on the website where they stated it would be. This bureaucratic hurdle appears to have rendered this exception useless, which is disappointing considering the TSA pushed for mDLs to be used first in their context.

Google ID Pass

Another exception appears to bypass state issued waivers, Google Wallet’s “ID Pass”. If a state partnered with Google to issue mDLs, or if you have a passport, then that is acceptable to TSA. This is a large leap in terms of reach of the mDL ecosystem expanding past state scrutiny to partnering directly with a private company to bring acceptable forms of ID for TSA. There’s much to be said on our worries with digital ID and the rapid expansion of them outside of the airport context. This is another gateway that highlights how ID is being shaped and accepted in the digital sense.

Both with ID Pass and mDLs, the presentation flow allows for you to tap with your phone without unlocking it. Which is a bonus, but it is not clear if TSA has the tech to read these IDs at all airports nationwide and it is still encouraged to bring a physical ID for additional verification.

A lot of the privilege dynamics of flying appear through types of ID you can obtain, whether your shoes stay on, how long you wait in line, etc. This is mostly tied to how much you can spend on traveling and how much preliminary information you establish with TSA ahead of time. The end result is that less wealthy people are subjected to the most security mechanisms at the security gate. For now, you can technically still fly without a REAL ID, but that means being subject to additional screening to verify who you are.

REAL ID enforcement has some leg room for those who do not want or can’t get a REAL ID. But the progression of digital ID is something we are keeping watch of that continues to be presented as the solution to worries of fraud and forgery. Governments and private corporations alike are pushing major efforts for rapid digital ID deployments and more frequent presentation of one’s ID attributes. Your government ID is one of the narrowest, static verifications of who you are as a person. Making sure that information is not used to create a centralized system of information was as important yesterday with REAL ID as it is today with digital IDs.

Lawmakers and regulators around the world have been prolific with passing legislation restricting freedom of expression and privacy for LGBTQ+ individuals and fueling offline intolerance. Online platforms are also complicit in this pervasive ecosystem by censoring pro-LGBTQ+ speech, forcing LGBTQ+ individuals to self-censor or turn to VPNs to avoid being profiled, harassed, doxxed, or criminally prosecuted.  

The fight for the safety and rights of LGBTQ+ people is not just a fight for visibility online (and offline)—it’s a fight for survival. This International Day Against Homophobia, Biphobia, and Transphobia, we’re sharing four essential tips for LGBTQ+ people to stay safe online.

Using Secure Messaging Services For Every Communication 

All of us, at least occasionally, need to send a message that’s safe from prying eyes. This is especially true for people who face consequences should their gender or sexual identity be revealed without their consent.

To protect your communications from being seen by others, install an encrypted messenger app such as Signal (for iOS or Android). Turn on disappearing messages, and consider shortening the amount of time messages are kept in the app if you are actually attending an event. If you have a burner device with you, be sure to save the numbers for emergency contacts.

Don’t wait until something sensitive arises: make these apps your default for all communications. As a side benefit, the messages and images sent to family and friends in group chats will be safe from being viewed by automated and human scans on services like Telegram and Facebook Messenger. 

Consider The Content You Post On Social Media 

Our decision to send messages, take pictures, and interact with online content has a real offline impact. And whilst we cannot control every circumstance, we can think about how our social media behaviour impacts those closest to us and those in our proximity, especially if these people might need extra protection around their identities. 

Talk with your friends about the potentially sensitive data you reveal about each other online. Even if you don’t have a social media account, or if you untag yourself from posts, friends can still unintentionally identify you, report your location, and make their connections to you public. This works in the offline world too, such as sharing precautions with organizers and fellow protesters when going to a demonstration, and discussing ahead of time how you can safely document and post the event online without exposing those in attendance to harm.

If you are organizing online or conversing on potentially sensitive issues, choose platforms that limit the amount of information collected and tracking undertaken. We know this is not always possible as perhaps people cannot access different applications. In this scenario, think about how you can protect your community on the platform you currently engage on. For example, if you currently use Facebook for organizing, work with others to keep your groups as private and secure as possible.

Create Incident Response Plans

Developing a plan for if or when something bad happens is a good practice for anyone, but especially for LGBTQ+ people who face increased risk online. Since many threats are social in nature, such as doxxing or networked harassment, it’s important to strategize with your allies around what to do in the event of such things happening. Doing so before an incident occurs is much easier than when you’re presently facing a crisis.

Only you and your allies can decide what belongs on such a plan, but some strategies might be: 

• Isolating the impacted areas, such as shutting down social media accounts and turning off affected devices
• Notifying others who may be affected
• Switching communications to a predetermined more secure alternative
• Noting behaviors of suspected threats and documenting these 
• Outsourcing tasks to someone further from the affected circle who is already aware of this potential responsibility.

Consider Your Safety When Attending and Protests 

Given the increase in targeted harassment and vandalism towards LGBTQ+ people, it’s important to consider counterprotesters showing up at various events. Since the boundaries between events like pride parades and protest might be blurred, precautions are necessary. Our general guide for attending a protest covers the basics for protecting your smartphone and laptop, as well as providing guidance on how to communicate and share information responsibly. We also have a handy printable version available here.

This includes:

• Removing biometric device unlock like fingerprint or FaceID to prevent police officers from physically forcing you to unlock your device with your fingerprint or face. You can password-protect your phone instead.
• Logging out of accounts and uninstalling apps or disabling app notifications to avoid app activity in precarious legal contexts from being used against you, such as using queer dating apps in places where homosexuality is illegal. 
• Turning off location services on your devices to avoid your location history from being used to identify your device’s comings and goings. For further protections, you can disable GPS, Bluetooth, Wi-Fi, and phone signals when planning to attend a protest.

LGBTQ+ Rights For Every Day 

Consider your digital safety like you would any aspect of bodily autonomy and self determination—only you get to decide what aspects of yourself you share with others, how you present to the world, and what things you keep private. With a bit of care, you can maintain privacy, safety, and pride in doing so. 

And in the meantime, we’re fighting to ensure that the internet can be a safe (and fun!) place for all LGBTQ+ people. Now more than ever, it’s essential for allies, advocates, and marginalized communities to push back against these dangerous laws and ensure that the internet remains a space where all voices can be heard, free from discrimination and censorship.

From Hackaday.com, this is a neural network simulation of a pet squid.

Autonomous Behavior:

• The squid moves autonomously, making decisions based on his current state (hunger, sleepiness, etc.).
• Implements a vision cone for food detection, simulating realistic foraging behavior.
• Neural network can make decisions and form associations.
• Weights are analysed, tweaked and trained by Hebbian learning algorithm.
• Experiences from short-term and long-term memory can influence decision-making.
• Squid can create new neurons in response to his environment (Neurogenesis) ...

This week, the U.S. House Ways and Means Committee moved forward with a proposal that would allow the Secretary of the Treasury to strip any U.S. nonprofit of its tax-exempt status by unilaterally determining the organization is a “Terrorist Supporting Organization.” This proposal, which places nearly unlimited discretion in the hands of the executive branch to target organizations it disagrees with, poses an existential threat to nonprofits across the U.S. 

This proposal, added to the House’s budget reconciliation bill, is an exact copy of a House-passed bill that EFF and hundreds of nonprofits across the country strongly opposed last fall. Thankfully, the Senate rejected that bill, and we urge the House to do the same when the budget reconciliation bill comes up for a vote on the House floor. 

The goal of this proposal is not to stop the spread of or support for terrorism; the U.S. already has myriad other laws that do that, including existing tax code section 501(p), which allows the government to revoke the tax status of designated “Terrorist Organizations.” Instead, this proposal is designed to inhibit free speech by discouraging nonprofits from working with and advocating on behalf of disadvantaged individuals and groups, like Venezuelans or Palestinians, who may be associated, even completely incidentally, with any group the U.S. deems a terrorist organization. And depending on what future groups this administration decides to label as terrorist organizations, it could also threaten those advocating for racial justice, LGBTQ rights, immigrant communities, climate action, human rights, and other issues opposed by this administration. 

On top of its threats to free speech, the language lacks due process protections for targeted nonprofit organizations. In addition to placing sole authority in the hands of the Treasury Secretary, the bill does not require the Treasury Secretary to disclose the reasons for or evidence supporting a “Terrorist Supporting Organization” designation. This, combined with only providing an after-the-fact administrative or judicial appeals process, would place a nearly insurmountable burden on any nonprofit to prove a negative—that they are not a terrorist supporting organization—instead of placing the burden where it should be, on the government. 

As laid out in letter led by ACLU and signed by over 350 diverse nonprofits, this bill would provide the executive branch with: 

“the authority to target its political opponents and use the fear of crippling legal fees, the stigma of the designation, and donors fleeing controversy to stifle dissent and chill speech and advocacy. And while the broadest applications of this authority may not ultimately hold up in court, the potential reputational and financial cost of fending off an investigation and litigating a wrongful designation could functionally mean the end of a targeted nonprofit before it ever has its day in court.” 

Current tax law makes it a crime for the President and other high-level officials to order IRS investigations over policy disagreements. This proposal creates a loophole to this rule that could chill nonprofits for years to come. 

There is no question that nonprofits and educational institutions – along with many other groups and individuals – are under threat from this administration. If passed, future administrations, regardless of party affiliation, could weaponize the powers in this bill against nonprofits of all kinds. We urge the House to vote down this proposal. 

“MIT Sloan was my first and only choice,” says MIT graduate student David Brown. After receiving his BS in chemical engineering at the U.S. Military Academy at West Point, Brown spent eight years as a helicopter pilot in the U.S. Army, serving as a platoon leader and troop commander. 

Now in the final year of his MBA, Brown has co-founded a climate tech company — Helix Carbon — with Ariel Furst, an MIT assistant professor in the Department of Chemical Engineering, and Evan Haas MBA ’24, SM ’24. Their goal: erase the carbon footprint of tough-to-decarbonize industries like ironmaking, polyurethanes, and olefins by generating competitively-priced, carbon-neutral fuels directly from waste carbon dioxide (CO2). It’s an ambitious project; they’re looking to scale the company large enough to have a gigaton per year impact on CO2 emissions. They have lab space off campus, and after graduation, Brown will be taking a full-time job as chief operating officer.

“What I loved about the Army was that I felt every day that the work I was doing was important or impactful in some way. I wanted that to continue, and felt the best way to have the greatest possible positive impact was to use my operational skills learned from the military to help close the gap between the lab and impact in the market.”

The following photo essay provides a snapshot of what a typical day for Brown has been like as an MIT student.

The Knight Science Journalism Program (KSJ) at MIT has announced that Usha Lee McFarling, national science correspondent for STAT and former KSJ Fellow, will be joining the team in August as their next director.

As director, McFarling will play a central role in helping to manage KSJ — an elite mid-career fellowship program that brings prominent science journalists from around the world for 10 months of study and intellectual exploration at MIT, Harvard University, and other institutions in the Boston area.

“I’m eager to take the helm during this critical time for science journalism, a time when journalism is under attack both politically and economically and misinformation — especially in areas of science and health — is rife,” says McFarling. “My goal is for the program to find even more ways to support our field and its practitioners as they carry on their important work.”

McFarling is a veteran science writer, most recently working for STAT News. She previously reported for the Los Angeles Times, The Boston Globe, Knight Ridder Washington Bureau, and the San Antonio Light, and was a Knight Science Journalism Fellow in 1992-93. McFarling graduated from Brown University with a degree in biology in 1988 and later earned a master’s degree in biological psychology from the University of California at Berkeley.

Her work on the diseased state of the world’s oceans earned the 2007 Pulitzer Prize for explanatory journalism and a Polk Award, among others. Her coverage of health disparities at STAT has earned an Edward R. Murrow award, and awards from the Association of Health Care Journalists, and the Asian American Journalists Association. In 2024, she was awarded the Victor Cohn prize for excellence in medical science reporting and the Bernard Lo, MD award in bioethics.

McFarling will succeed director Deborah Blum, who served as director for 10 years. Blum, also a Pulitzer-prize winning journalist and the bestselling author of six books, is retiring to return to a full-time writing career. She will join the board of Undark, a magazine she helped found while at KSJ, and continue as a board member of the Council for the Advancement of Science Writing and the Burroughs Wellcome Fund, among others.

“It’s been an honor to serve as director of the Knight Science Journalism program for the past 10 years and a pleasure to be able to support the important work that science journalists do,” Blum says. “And I know that under the direction of Usha McFarling — who brings such talent and intelligence to the job — that KSJ will continue to grow and thrive in all the best ways.”

This is a weird story:

U.S. energy officials are reassessing the risk posed by Chinese-made devices that play a critical role in renewable energy infrastructure after unexplained communication equipment was found inside some of them, two people familiar with the matter said.

[…]

Over the past nine months, undocumented communication devices, including cellular radios, have also been found in some batteries from multiple Chinese suppliers, one of them said.

Reuters was unable to determine how many solar power inverters and batteries they have looked at...

The EPA administrator might have provided a road map to revoking the critical scientific finding in a rule that repeals the power plant climate regulation.

David Richardson told staff in a private meeting that FEMA’s role won’t change much in 2025, even as the administration considers reducing aid to states.

Jonathan Pershing previously served as a climate envoy for the State Department.

The proposed regulation would force employers to provide workers with water and a cool place to rest when temperatures climb.

“Ignoring climate change’s destabilizing effects on the economy is not an option,” the lawmakers wrote.

The deal entails delivery of carbon credits generated from projects that sequester carbon dioxide by planting trees or restoring degraded land.

The European Commission will propose the measure by the end of 2026 as part of a scheme to make data centers more sustainable.

The European Commission has proposed amendments to environmental, social and governance legislation amid complaints that the rules pose too great a regulatory burden on business.

The women have created what are known as dream maps, showing their villages in their ideal states.

Within the next decade, generative AI could join computers and electricity as one of the most transformational technologies in history, with all of the promise and peril that implies. Governments’ responses to GenAI—including new legal precedents—need to thoughtfully address real-world harms without destroying the public benefits GenAI can offer. Unfortunately, the U.S. Copyright Office’s rushed draft report on AI training misses the mark.

The Report Bungles Fair Use

Released amidst a set of controversial job terminations, the Copyright Office’s report covers a wide range of issues with varying degrees of nuance. But on the core legal question—whether using copyrighted works to train GenAI is a fair use—it stumbles badly. The report misapplies long-settled fair use principles and ultimately puts a thumb on the scale in favor of copyright owners at the expense of creativity and innovation.

To work effectively, today’s GenAI systems need to be trained on very large collections of human-created works—probably millions of them. At this scale, locating copyright holders and getting their permission is daunting for even the biggest and wealthiest AI companies, and impossible for smaller competitors. If training makes fair use of copyrighted works, however, then no permission is needed.

Right now, courts are considering dozens of lawsuits that raise the question of fair use for GenAI training. Federal District Judge Vince Chhabria is poised to rule on this question, after hearing oral arguments in Kadrey v. Meta Platforms. The Third Circuit Court of Appeals is expected to consider a similar fair use issue in Thomson Reuters v. Ross Intelligence. Courts are well-equipped to resolve this pivotal issue by applying existing law to specific uses and AI technologies. 

Courts Should Reject the Copyright Office’s Fair Use Analysis

The report’s fair use discussion contains some fundamental errors that place a thumb on the scale in favor of rightsholders. Though the report is non-binding, it could influence courts, including in cases like Kadrey, where plaintiffs have already filed a copy of the report and urged the court to defer to its analysis.   

Courts need only accept the Copyright Office’s draft conclusions, however, if they are persuasive. They are not.   

The Office’s fair use analysis is not one the courts should follow. It repeatedly conflates the use of works for training models—a necessary step in the process of building a GenAI model—with the use of the model to create substantially similar works. It also misapplies basic fair use principles and embraces a novel theory of market harm that has never been endorsed by any court.

The first problem is the Copyright Office’s transformative use analysis. Highly transformative uses—those that serve a different purpose than that of the original work—are very likely to be fair. Courts routinely hold that using copyrighted works to build new software and technology—including search engines, video games, and mobile apps—is a highly transformative use because it serves a new and distinct purpose. Here, the original works were created for various purposes and using them to train large language models is surely very different.

The report attempts to sidestep that conclusion by repeatedly ignoring the actual use in question—training —and focusing instead on how the model may be ultimately used. If the model is ultimately used primarily to create a class of works that are similar to the original works on which it was trained, the Office argues, then the intermediate copying can’t be considered transformative. This fundamentally misunderstands transformative use, which should turn on whether a model itself is a new creation with its own distinct purpose, not whether any of its potential uses might affect demand for a work on which it was trained—a dubious standard that runs contrary to decades of precedent.

The Copyright Office’s transformative use analysis also suggests that the fair use analysis should consider whether works were obtained in “bad faith,” and whether developers respected the right “to control” the use of copyrighted works.  But the Supreme Court is skeptical that bad faith has any role to play in the fair use analysis and has made clear that fair use is not a privilege reserved for the well-behaved. And rightsholders don’t have the right to control fair uses—that’s kind of the point.

Finally, the Office adopts a novel and badly misguided theory of “market harm.” Traditionally, the fair use analysis requires courts to consider the effects of the use on the market for the work in question. The Copyright Office suggests instead that courts should consider overall effects of the use of the models to produce generally similar works. By this logic, if a model was trained on a Bridgerton novel—among millions of other works—and was later used by a third party to produce romance novels, that might harm series author Julia Quinn’s bottom line.

This market dilution theory has four fundamental problems. First, like the transformative use analysis, it conflates training with outputs. Second, it’s not supported by any relevant precedent. Third, it’s based entirely on speculation that Bridgerton fans will buy random “romance novels” instead of works produced by a bestselling author they know and love.  This relies on breathtaking assumptions that lack evidence, including that all works in the same genre are good substitutes for each other—regardless of their quality, originality, or acclaim. Lastly, even if competition from other, unique works might reduce sales, it isn’t the type of market harm that weighs against fair use.

Nor is lost revenue from licenses for fair uses a type of market harm that the law should recognize. Prioritizing private licensing market “solutions” over user rights would dramatically expand the market power of major media companies and chill the creativity and innovation that copyright is intended to promote. Indeed, the fair use doctrine exists in part to create breathing room for technological innovation, from the phonograph record to the videocassette recorder to the internet itself. Without fair use, crushing copyright liability could stunt the development of AI technology.

We’re still digesting this report, but our initial review suggests that, on balance, the Copyright Office’s approach to fair use for GenAI training isn’t a dispassionate report on how existing copyright law applies to this new and revolutionary technology. It’s a policy judgment about the value of GenAI technology for future creativity, by an office that has no business making new, free-floating policy decisions.

The courts should not follow the Copyright Office’s speculations about GenAI. They should follow precedent.

A protein located in the wrong part of a cell can contribute to several diseases, such as Alzheimer’s, cystic fibrosis, and cancer. But there are about 70,000 different proteins and protein variants in a single human cell, and since scientists can typically only test for a handful in one experiment, it is extremely costly and time-consuming to identify proteins’ locations manually.

A new generation of computational techniques seeks to streamline the process using machine-learning models that often leverage datasets containing thousands of proteins and their locations, measured across multiple cell lines. One of the largest such datasets is the Human Protein Atlas, which catalogs the subcellular behavior of over 13,000 proteins in more than 40 cell lines. But as enormous as it is, the Human Protein Atlas has only explored about 0.25 percent of all possible pairings of all proteins and cell lines within the database.

Now, researchers from MIT, Harvard University, and the Broad Institute of MIT and Harvard have developed a new computational approach that can efficiently explore the remaining uncharted space. Their method can predict the location of any protein in any human cell line, even when both protein and cell have never been tested before.

Their technique goes one step further than many AI-based methods by localizing a protein at the single-cell level, rather than as an averaged estimate across all the cells of a specific type. This single-cell localization could pinpoint a protein’s location in a specific cancer cell after treatment, for instance.

The researchers combined a protein language model with a special type of computer vision model to capture rich details about a protein and cell. In the end, the user receives an image of a cell with a highlighted portion indicating the model’s prediction of where the protein is located. Since a protein’s localization is indicative of its functional status, this technique could help researchers and clinicians more efficiently diagnose diseases or identify drug targets, while also enabling biologists to better understand how complex biological processes are related to protein localization.

“You could do these protein-localization experiments on a computer without having to touch any lab bench, hopefully saving yourself months of effort. While you would still need to verify the prediction, this technique could act like an initial screening of what to test for experimentally,” says Yitong Tseo, a graduate student in MIT’s Computational and Systems Biology program and co-lead author of a paper on this research.

Tseo is joined on the paper by co-lead author Xinyi Zhang, a graduate student in the Department of Electrical Engineering and Computer Science (EECS) and the Eric and Wendy Schmidt Center at the Broad Institute; Yunhao Bai of the Broad Institute; and senior authors Fei Chen, an assistant professor at Harvard and a member of the Broad Institute, and Caroline Uhler, the Andrew and Erna Viterbi Professor of Engineering in EECS and the MIT Institute for Data, Systems, and Society (IDSS), who is also director of the Eric and Wendy Schmidt Center and a researcher at MIT’s Laboratory for Information and Decision Systems (LIDS). The research appears today in Nature Methods.

Collaborating models

Many existing protein prediction models can only make predictions based on the protein and cell data on which they were trained or are unable to pinpoint a protein’s location within a single cell.

To overcome these limitations, the researchers created a two-part method for prediction of unseen proteins’ subcellular location, called PUPS.

The first part utilizes a protein sequence model to capture the localization-determining properties of a protein and its 3D structure based on the chain of  amino acids that forms it.

The second part incorporates an image inpainting model, which is designed to fill in missing parts of an image. This computer vision model looks at three stained images of a cell to gather information about the state of that cell, such as its type, individual features, and whether it is under stress.

PUPS joins the representations created by each model to predict where the protein is located within a single cell, using an image decoder to output a highlighted image that shows the predicted location.

“Different cells within a cell line exhibit different characteristics, and our model is able to understand that nuance,” Tseo says.

A user inputs the sequence of amino acids that form the protein and three cell stain images — one for the nucleus, one for the microtubules, and one for the endoplasmic reticulum. Then PUPS does the rest.

A deeper understanding

The researchers employed a few tricks during the training process to teach PUPS how to combine information from each model in such a way that it can make an educated guess on the protein’s location, even if it hasn’t seen that protein before.

For instance, they assign the model a secondary task during training: to explicitly name the compartment of localization, like the cell nucleus. This is done alongside the primary inpainting task to help the model learn more effectively.

A good analogy might be a teacher who asks their students to draw all the parts of a flower in addition to writing their names. This extra step was found to help the model improve its general understanding of the possible cell compartments.

In addition, the fact that PUPS is trained on proteins and cell lines at the same time helps it develop a deeper understanding of where in a cell image proteins tend to localize.

PUPS can even understand, on its own, how different parts of a protein’s sequence contribute separately to its overall localization.

“Most other methods usually require you to have a stain of the protein first, so you’ve already seen it in your training data. Our approach is unique in that it can generalize across proteins and cell lines at the same time,” Zhang says.

Because PUPS can generalize to unseen proteins, it can capture changes in localization driven by unique protein mutations that aren’t included in the Human Protein Atlas.

The researchers verified that PUPS could predict the subcellular location of new proteins in unseen cell lines by conducting lab experiments and comparing the results. In addition, when compared to a baseline AI method, PUPS exhibited on average less prediction error across the proteins they tested.

In the future, the researchers want to enhance PUPS so the model can understand protein-protein interactions and make localization predictions for multiple proteins within a cell. In the longer term, they want to enable PUPS to make predictions in terms of living human tissue, rather than cultured cells.

This research is funded by the Eric and Wendy Schmidt Center at the Broad Institute, the National Institutes of Health, the National Science Foundation, the Burroughs Welcome Fund, the Searle Scholars Foundation, the Harvard Stem Cell Institute, the Merkin Institute, the Office of Naval Research, and the Department of Energy.