Connect with us

Technology

Researchers Crack Open the ‘Black Box’ of Protein AI Models

The approach could accelerate drug target identification, vaccine research, and new biological discoveries.

Published

on

illustration ai research jpeg
Illustrated image

For years, artificial intelligence models that predict protein structures and functions have been critical tools in drug discovery, vaccine development, and therapeutic antibody design. But while these protein language models (PLMs), often built on large language models (LLMs), deliver impressively accurate predictions, researchers have been unable to see how the models arrive at those decisions — until now.

In a study published this week in the Proceedings of the National Academy of Sciences (PNAS), a team of MIT researchers unveiled a novel method to interpret the inner workings of these black-box models. By shedding light on the features that influence predictions, the approach could accelerate drug target identification, vaccine research, and new biological discoveries.

Cracking the protein ‘black box’

“Protein language models have been widely used for many biological applications, but there’s always been a missing piece: explainability,” said Bonnie Berger, Simons Professor of Mathematics and head of the Computation and Biology group in MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). In a media statement, she explained, “Our work has broad implications for enhanced explainability in downstream tasks that rely on these representations. Additionally, identifying features that protein language models track has the potential to reveal novel biological insights.”

The study was led by MIT graduate student Onkar Gujral, with contributions from Mihir Bafna, also a graduate student, and Eric Alm, professor of biological engineering at MIT.

From AlphaFold to explainability

Protein modelling took off in 2018 when Berger and then-graduate student Tristan Bepler introduced the first protein language model. These models, much like ChatGPT processes words, analyze amino acid sequences to predict protein structure and function. Their innovations paved the way for powerful systems like AlphaFold, ESM2, and OmegaFold, transforming the fields of bioinformatics and molecular biology.

Yet, despite their predictive power, researchers remained in the dark about why a model reached certain conclusions. “We would get out some prediction at the end, but we had absolutely no idea what was happening in the individual components of this black box,” Berger noted.

The sparse autoencoder approach

To address this challenge, the MIT team employed a technique called a sparse autoencoder — an algorithm originally used to interpret LLMs. Sparse autoencoders expand the representation of a protein across thousands of neural nodes, making it easier to distinguish which specific features influence the prediction.

“In a sparse representation, the neurons lighting up are doing so in a more meaningful manner,” explained Gujral in a media statement. “Before the sparse representations are created, the networks pack information so tightly together that it’s hard to interpret the neurons.”

By analyzing these expanded representations using AI assistance from Claude, the researchers could link specific nodes to biological features such as protein families, molecular functions, or even their location in a cell. For instance, one node could be identified as signalling proteins involved in transmembrane ion transport.

Implications for drug discovery and biology

This new transparency could be transformational for drug design and vaccine development, allowing scientists to select the most reliable models for specific biomedical tasks. Moreover, the study suggests that as AI models become more powerful, they could reveal previously undiscovered biological patterns.

“Understanding what features protein models encode means researchers can fine-tune inputs, select optimal models, and potentially even uncover new biological insights from the models themselves,” Gujral said. “At some point, when these models get more powerful, you could learn more biology than you already know just from opening up the models.”

Technology

What Nitrogen-Fixing Microbes Could Teach Us About Cleaner Fertiliser

MIT research into nitrogen-fixing enzymes reveals how microbes efficiently convert atmospheric nitrogen into ammonia, offering clues for developing cleaner and more energy-efficient fertiliser production.

Published

on

Alt text: Nitrogen fertiliser in a scoop over soil and organic compost
Nitrogen fertiliser being handled alongside soil and organic compost. Representational image. Image credit: Kaboompics/Pexels.

Nitrogen makes up nearly four-fifths of the atmosphere, but turning that abundant gas into a fertilizer plants can use is not easy. The nitrogen molecule, N₂, is held together by one of the strongest chemical bonds in nature.

However, some microbes, using enzymes called nitrogenases, convert atmospheric nitrogen into ammonia, which can then be used to build proteins and other essential molecules.

Now, studies from researchers at the Massachusetts Institute of Technology (MIT) are helping explain why one class of these enzymes works so well. The findings could eventually guide the design of synthetic catalysts for producing ammonia with less energy, a possibility that could matter for the future of fertiliser manufacturing.

The Tiny Chemical Trick Behind Nitrogen Fixation

Nitrogen-fixing enzymes, called nitrogenases, come in three main types depending on the metal at their active site: molybdenum, vanadium, or iron. Among these, the molybdenum-based version is the most efficient at converting nitrogen gas into ammonia. But scientists have long wondered why this is the case, especially since iron is thought to be the main site where nitrogen actually binds.

To investigate this, MIT researchers studied simplified versions of the enzyme using iron–sulfur clusters. They found that larger metal atoms like molybdenum and tungsten helped the cluster hold N₂ gas more strongly, while smaller metals such as iron, vanadium, and chromium did not show the same effect.

A follow-up study suggested an explanation: molybdenum may not need to directly bind nitrogen at all. Instead, it can influence nearby iron atoms through electronic interactions, making it easier for iron to transfer electrons to nitrogen. This electron transfer is a key step in weakening the strong nitrogen bond so it can eventually be converted into ammonia. In simple terms, it means that molybdenum seems to assist iron in doing the hardest part of the reaction.

From Microbial Enzymes to Cleaner Ammonia

Ammonia is the starting point for most N₂ fertilisers, including urea. Industrial ammonia production relies mainly on the Haber-Bosch process, which requires substantial energy. That makes ammonia production an important target for efforts to decarbonise the fertiliser sector.

Hands holding nitrogen fertiliser granules for agricultural use
Nitrogen fertiliser granules held in a farmer’s hands. Representational image. Image credit: Kashif Shah/Pexels.

The MIT findings offer a design principle that catalysts may be made more effective by getting different metals to cooperate electronically. If such principles can eventually be translated into robust synthetic catalysts, they could help researchers explore ammonia production under less energy-intensive conditions.

Why This Matters to India: Green Ammonia

For India, the question is particularly relevant because ammonia sits at the centre of the fertiliser system. Producing it through the conventional Haber–Bosch process is highly energy-intensive, requiring high temperatures and pressures and accounting for a significant share of global industrial energy use, largely supplied by fossil fuels. This not only adds to production costs but also links fertiliser prices to energy markets. The country has also faced periodic fertiliser shortages and import dependence, making efficient and lower-energy ammonia production strategically important.

India is already building a policy framework around green ammonia. The Ministry of New and Renewable Energy issued a Green Ammonia Standard for India in February 2026, while projects for green ammonia production are being developed across states including Karnataka, Tamil Nadu, Odisha, Rajasthan and Andhra Pradesh.

That makes the MIT research relevant beyond the laboratory.  The bigger challenge is still ahead. Nitrogenase is an extraordinarily complex biological system, and reproducing its efficiency, stability and selectivity in an industrial catalyst remains difficult.

But microbes have already demonstrated that atmospheric nitrogen does not have to remain chemically out of reach. The real challenge now is whether scientists can translate this biological solution into a process that is efficient, stable, and scalable enough for industrial use.

Continue Reading

Technology

Can Integrated Clean Energy Reshape India’s Steel Industry?

India’s steel industry is expanding rapidly, but reducing its carbon footprint remains a major challenge. A new study suggests that integrating renewable electricity with green hydrogen could make low-carbon steel more affordable by cutting energy waste and limiting cost increases. The findings offer fresh insights into how smarter energy planning could support India’s green steel ambitions.

Published

on

India's green steel industry transformation underway
India's steel industry is exploring integrated renewable electricity and green hydrogen to reduce carbon emissions while maintaining competitiveness. Representational image. Image credit: Kateryna Babaieva/Pexels

India’s steel industry is at a pivotal moment. As the world’s second-largest crude steel producer, India plans to expand production capacity to 300 million tonnes by 2030-31. Steel will be central to the country’s infrastructure, housing, renewable energy and manufacturing ambitions. But the sector is also one of India’s biggest climate challenges.

Unlike the power sector, steel cannot be decarbonised simply by switching to renewable electricity. Its production relies on high-temperature processes and chemical reactions that still depend largely on coal. According to India’s draft National Steel Policy 2025, cited by Reuters, the steel sector contributes 10–12% of the country’s greenhouse gas emissions. Producing one tonne of finished steel emits 2.65 tonnes of CO₂, well above the global average of 2 tonnes.

A recent analysis by climate-tech think tank TransitionZero suggests the solution may lie in rethinking how clean energy is used. Rather than viewing renewable electricity and green hydrogen as separate technologies, the study explores whether integrating the two could make steel production cleaner without substantially increasing costs.

The challenge of Replacing Coal

Much of the push to decarbonise steel has centred on green hydrogen, which can replace coal or natural gas in direct reduced iron (DRI) production. Recognising its potential, India launched the National Green Hydrogen Mission, targeting 5 million metric tonnes of green hydrogen annually by 2030.

However, green hydrogen remains costly because its production requires large amounts of renewable electricity. A 2021 study by the Council on Energy, Environment and Water (CEEW) found that steel produced entirely with green hydrogen is unlikely to become commercially competitive before 2040 unless production costs fall significantly. The challenge, therefore, is not just developing cleaner fuels, but using clean energy more efficiently.

Steel Industry: Rethinking How Clean Energy is Used

Solar and wind farms generate electricity for the grid, while hydrogen producers source renewable power independently. TransitionZero argues that this approach overlooks a significant opportunity for steel industry. The researchers simulated how India’s projected electricity grid would operate in 2030, analysing every hour of the year to identify when surplus renewable electricity could be used to produce green hydrogen instead of being wasted. 

Steel industry in india
Source: TransitionZero estimates for 2030 steel industry, based on capacity and production data from GEM and India’s Ministry of Steel.

Solar power generation often exceeds demand during the day, leaving the grid unable to absorb all the electricity produced. Instead of curtailing this surplus renewable energy, the report proposes using it to power electrolysers that produce green hydrogen. The hydrogen can then be stored and used in steel production when renewable electricity is less abundant.

According to the analysis, this integrated approach could reduce renewable energy curtailment by up to 90 per cent while increasing steel production costs by only around 3 per cent. Steel plants sourcing 70 per cent carbon-free electricity every hour and replacing 20 per cent of natural gas with green hydrogen could significantly cut emissions without substantially raising costs. The findings suggest that better coordination between renewable electricity and hydrogen may be as important as the technologies themselves.

Building on Evidence

The idea of combining multiple technologies to decarbonise steel industry is not new. The International Energy Agency identifies hydrogen-based direct reduced iron, electric arc furnaces, steel recycling and energy efficiency as key pathways to achieving net-zero steel production. Similarly, the Council on Energy, Environment and Water (CEEW) has argued that India should prioritise expanding renewable electricity while gradually introducing green hydrogen as costs become more competitive.

TransitionZero builds on these recommendations by focusing on how these technologies can work together. Rather than treating renewable electricity and green hydrogen as separate solutions, the study shows that integrating them can improve energy use, reduce costs and lower emissions. The findings underscore a broader shift in industrial decarbonisation—from adopting cleaner technologies to designing smarter, more integrated energy systems should be used in steel industry.

A Question of Competitiveness, Not Just Climate

Although India consumes most of the steel it produces domestically, exporters are preparing for stricter environmental standards in international markets. The European Union’s Carbon Border Adjustment Mechanism (CBAM), which will gradually impose carbon costs on imported steel and other emissions-intensive products, could increase costs for producers with high carbon footprints.

Reducing emissions is therefore no longer solely about meeting climate targets. It is increasingly linked to maintaining access to export markets and improving industrial competitiveness.

Indian steelmakers have already begun responding. Companies including Tata Steel, JSW Steel and ArcelorMittal Nippon Steel India are investing in renewable energy, exploring hydrogen-based technologies and testing lower-carbon production processes. These projects remain at an early stage, but they indicate that the steel industry’s transition has already begun.

Planning the Transition Of Technology

India has no shortage of technologies capable of reducing emissions from steel production. Renewable electricity is expanding rapidly, hydrogen technologies are maturing and electric arc furnaces are becoming more efficient. The greater challenge lies in connecting these pieces into a coherent industrial strategy.

As India’s steel industry moves towards its 300-million-tonne ambition, success will depend less on efficiently designing energy systems that work together. Cleaner steel industry may ultimately depend not on one revolutionary technology, but on rethinking how India’s energy and industrial systems operate together.

Continue Reading

Technology

Global Experts Seek Treaty to Keep AI Out of Nuclear Decisions

Global experts are urging a treaty to keep artificial intelligence out of nuclear weapons decisions, warning that human judgment must remain central to global security.

Published

on

Military tank symbolizing concerns over artificial intelligence in nuclear weapons decision-making and autonomous warfare.
Military experts and global leaders are calling for international rules to ensure artificial intelligence never makes decisions on the use of nuclear weapons. Representational image. Image credit: Pexels

A coalition of Nobel laureates, artificial intelligence researchers, religious leaders and public figures has called for an international treaty to prevent Artificial Intelligence from controlling nuclear weapons, warning that decisions affecting millions of lives should never be left to autonomous systems.

The appeal comes through the Rome Declaration for an Unarmed and Disarming Peace, signed on July 16. At a time when militaries are rapidly adopting Artificial Intelligence for surveillance, intelligence and battlefield operations, the signatories say international rules have failed to keep pace with technological advances. They want governments to draw a clear line by prohibiting Artificial Intelligence from making the final decision on the use of nuclear weapons.

Concerns Over Faster Decisions

The declaration warns that Artificial Intelligence could dramatically shorten the time available for leaders to assess threats during a nuclear crisis. If computer systems analyse incoming data and recommend a response within seconds, decision-makers may have little opportunity to verify information, consult advisers or pursue diplomacy before acting.

Its authors point to historical incidents such as the Cuban Missile Crisis in 1962 and the 1983 Soviet nuclear false alarm, where human judgement and restraint prevented escalation. They argue that replacing this layer of caution with automated systems could increase the risk of unintended conflict, especially as current AI models remain vulnerable to errors, manipulated data and opaque decision-making.

Call for Global Rules

Along with keeping humans in charge of nuclear decisions, the declaration recommends independent security audits of nuclear command systems to protect them from AI-enabled cyberattacks. It also urges Artificial Intelligence developers to disclose the ethical safeguards built into their models and renews calls for international negotiations on nuclear disarmament. The declaration is not legally binding, and countries remain divided over regulating AI in military applications. Even so, its signatories hope it will build support for global rules before advances in Artificial Intelligence outstrip the international mechanisms meant to govern them.

Continue Reading

Trending