AI model identifies hundreds of promising plant proteins for sustainable products
Scientists from Leeds’ School of Food Science and Nutrition have developed an advanced AI process that can rapidly identify plant proteins capable of acting as emulsifiers from tens of millions of initial candidates. To date, it has already identified nearly 800. The discovery will cut years of costly trial-and-error research and bring the next generation of plant-based foods and cosmetics closer.
Emulsifiers are essential for binding oil and water into stable, homogenous mixtures in food, cosmetics, pharmaceuticals and industrial applications. Common uses include lotions, medicinal creams, sauces, ice creams, mayonnaise and paints. There is increasing interest in creating natural, sustainable alternatives to high carbon-footprint synthetic or animal-derived emulsifiers.
The research, published today in Communications Chemistry, was led by postdoctoral researcher Dr Simha Sridharan and supervised by Professor Anwesha Sarkar, both of the University’s Sarkar Lab. They worked alongside AI researchers at Leeds’ School of Food Science and Nutrition and in close collaboration with Dr Rik Sarkar, a machine learning expert at the University of Edinburgh.
Dr Sridharan said: “As we want to shift towards more sustainable, plant‑based ingredients, scientists face a major challenge: There are millions of potential plant proteins but testing them all to identify the right emulsifier is expensive and involves a time‑consuming trial-and-error approach. Until now, there has been no reliable way to predict which plant proteins are likely to behave as emulsifiers like animal proteins.”
There is growing interest among consumers for natural emulsifiers, in place of the commonly used animal-based emulsifiers such as milk proteins like caseins or whey. This new tool could help identify new possibilities, reduce years of testing, and accelerate the transition towards sustainable, plant-based food systems.
Dr Sridharan and Professor Anwesha Sarkar used a simulation model to understand how proteins attach between oil and water mixtures, which is crucial for them to act as an emulsifier. In collaboration with Dr Rik Sarkar, they applied machine learning to fingerprint specific segments of the protein that dictate their attachment behaviour. By combining machine learning and statistical physics, the team were able to screen plant proteins to identify those that will resemble the emulsification performance of animal proteins in a fraction of the time needed by conventional experiments.
Dr Rik Sarkar said: “Emulsfiers often have a characteristic chemical structure called di-blocks. We were able to model this structure mathematically for plant proteins. Using machine learning based on features obtained from statistical physics simulations, we can predict which plant proteins are most likely to work best as natural emulsifiers.”
The team believe their discovery could be of interest to food and cosmetics companies developing plant‑based and sustainable products.
Professor Anwesha Sarkar, who is also co-director of the National Alternative Protein Innovation Centre (NAPIC) based at Leeds, added: “The model identified nearly 800 plant proteins that could potentially act as emulsifiers, many of which had never previously been considered for this purpose. We then tested several commercially available proteins and found the results matched the model's predictions, with proteins from peas and potatoes proving effective. This shows how AI could help researchers find promising new ingredients much faster than before.”
Further information
“Data-driven pipeline enables discovery of plant protein surfactants” will be published in Communications Chemistry at 10:00am GMT on Thursday 3rd September.
For media enquries please contact University of Leeds media officer Morgan Buswell via email on pressoffice@leeds.ac.uk.
University of Leeds
The University of Leeds is one of the largest higher education institutions in the UK, with more than 40,000 students from about 140 different countries. We are renowned globally for the quality of our teaching and research.
We are a values-driven university, and we harness our expertise in research and education to help shape a better future for humanity, working through collaboration to tackle inequalities, achieve societal impact and drive change.
The University is a member of the Russell Group of research-intensive universities, and is a major partner in the Alan Turing, Rosalind Franklin and Royce Institutes www.leeds.ac.uk
Follow University of Leeds or tag us in to coverage: Bluesky | Facebook | LinkedIn | Instagram
Journal
Communications Chemistry
Method of Research
Computational simulation/modeling
Subject of Research
Not applicable
Article Title
Data-driven pipeline enables discovery of plant protein surfactants
Article Publication Date
3-Sep-2026
Ancient Dujiangyan water wisdom inspires AI microfluidic chip to spot hidden sub-resistant bacteria
image:
Conventional MIC breakpoint-based classification may capture only the “tip of the iceberg,” leaving a hidden reservoir of sub-resistant bacteria unresolved that could evolve into overt resistance under sustained antibiotic pressure. To address this limitation, the research team developed DP-AST, an AI-assisted rapid antibiotic susceptibility testing platform inspired by the “six-four water diversion” principle of the Dujiangyan irrigation system. By translating ancient hydraulic wisdom into microfluidic chip design, DP-AST enables automated evaluation of bacterial growth activity and antibiotic concentration-effect curves, thereby supporting more precise susceptibility assessment and antibiotic treatment guidance.
view moreCredit: ©Science Bulletin
Antimicrobial resistance is one of the most urgent threats to global public health. In clinical practice, antibiotic treatment sometimes fails even when bacteria are classified as susceptible by conventional antimicrobial susceptibility testing. One important reason is that traditional tests mainly rely on the minimum inhibitory concentration, or MIC, and may overlook bacterial subpopulations that survive antibiotic exposure, including persistent, tolerant, and heteroresistant populations. These hidden populations can survive under drug pressure and may contribute to treatment failure and the later emergence of stable resistance.
To address this challenge, a research team led by Professor Bi-feng Liu developed a deep learning-based microfluidic rapid phenotypic AST system, named DP-AST. The system was inspired by the “six-four water diversion” principle of the ancient Dujiangyan irrigation system in China, which has long been known for its efficient and self-regulated water distribution. By translating this principle into microfluidic chip design, the team created a hand-driven concentration-gradient generator that can automatically produce multiple antibiotic concentrations in a portable format.
Unlike many existing rapid AST platforms that rely on colorimetric reagents, professional instruments, or high-end microscopes, DP-AST uses bacterial micro-enrichment area as a phenotypic readout. This design enables the system to calculate bacterial growth activity and generate concentration-effect curves, which describe how bacterial growth changes across different antibiotic concentrations. By combining MIC, growth activity, and concentration-effect curve analysis, DP-AST provides a more refined view of bacterial drug response and can identify sub-resistant bacterial populations that are difficult to detect using conventional MIC-based testing alone.
The platform also integrates smartphone-based signal acquisition with deep learning image analysis. Bacterial growth signals can be captured using a portable smartphone imaging setup, and the deep learning algorithm automatically analyzes the images and classifies the drug susceptibility results. This combination improves portability and supports low-instrumentation testing, which is especially important for resource-limited clinical settings where access to large laboratory instruments may be restricted.
Beyond platform development, the team further investigated the biological mechanisms underlying sub-resistance. Using integrated proteomic and transcriptomic analyses, they compared a sub-resistant Escherichia coli strain with a sensitive strain. Although the two strains showed highly similar genomic sequences, they displayed marked differences in gene and protein expression. The sub-resistant strain showed increased expression of several antibiotic response-related genes and proteins, including outer membrane protein OmpA and efflux system-related components, suggesting that expression-level regulation may help bacteria survive antibiotic pressure even without clear conventional resistance markers.
Further multi-omics analysis indicated that many of the transcriptional and proteomic changes in the sub-resistant strain were associated with intracellular acid regulation. The team experimentally identified gadE, a central regulator of acid tolerance in E. coli, as an important factor involved in bacterial survival under antibiotic stress. This finding suggests that acid tolerance-related regulatory networks may contribute to bacterial adaptation during antibiotic exposure and provides a new direction for studying sub-resistance mechanisms.
Overall, this study establishes a portable, AI-assisted microfluidic platform for rapid phenotypic antibiotic susceptibility testing. DP-AST can provide susceptibility results within 3 hours and, more importantly, enables multi-parameter profiling for the identification of hidden sub-resistant bacteria. The work offers a new technical strategy for more precise antibiotic treatment, improved resistance-risk assessment, and future studies on the early evolution of antimicrobial resistance.
Check for copycat bias in medical AI
Validating artificial intelligence performance in radiology studies with demographic subgroups
image:
Medical AI products may be forming and amplifying healthcare bias against demographic subgroups.
view moreCredit: Osaka Metropolitan University
The recent flood of Artificial Intelligence (AI) models and content has undoubtedly brought along an unwelcome bug from the human-made content it scrapes from: bias. AI has been noted to form, copy, and amplify harmful stereotypes of already marginalized groups. This phenomenon has far reaching effects as AI becomes a normal part of everyday life. Most worrying is that the prolific bias already circling healthcare settings may now be exacerbated by AI-driven decisions. These can have a direct impact on patient care and outcomes, underscoring the need for careful evaluation of potential biases.
Therefore, a research group led by Dr. Shannon L. Walston at Osaka Metropolitan University’s Graduate School of Medicine conducted a scoping review to identify studies validating commercially available radiology AI products and to note trends when reporting on sex, age, and ethnic demographic subgroups. The team collected 545 studies on 252 products with reported demographic subgroup data. Trends were mapped using a regression analysis.
Of the 545, only 77 studies validating 52 products were found to include demographic details and subgroup analysis results. When the Wilson Confidence Interval formula, which is used to calculate proportion, was applied to studies validating AI for tuberculosis detection, the researchers found that 67% of the reported datasets were at risk of being underpowered for sex subgroup analysis. This revealed that, despite the demand for improved demographics reporting to address the dangers of biased medical AI, performance reporting for demographic subgroups has not become more common.
This finding exposes the need for effective, transparent reporting to confirm the safe and unbiased performance of these medical AI products across all patient subgroups.
“This scoping review quantifies how fragmented the commercial validation landscape is, showing that reporting for both the demographics and per-subgroup performance is inadequate for estimating subgroup bias. This systemic problem requires effort from all stakeholders, from researchers to regulatory agencies, encouraging thorough reporting and commercial product validation to support physician and patient trust in medical AI products," stated Dr. Walston.
The findings were published in European Radiology.
###
About OMU
Established in Osaka as one of the largest public universities in Japan, Osaka Metropolitan University is committed to shaping the future of society through the “Convergence of Knowledge” and the promotion of world-class research. For more research news, visit https://www.omu.ac.jp/en/ and follow us on social media: X, Instagram, LinkedIn.
Journal
European Radiology
Method of Research
Systematic review
Subject of Research
People
Article Title
The current state of demographic subgroup reporting for commercially available AI for radiology: a scoping review
New AI BioDesign accelerator combines experimental biology and artificial intelligence to learn nature’s design rules
Researchers from the Allen Institute, the University of Washington, and Fred Hutch Cancer Center, supported by Fund for Science and Technology, will build open AI models, datasets, and tools to accelerate biological design for human health.
Allen Institute
image:
Sud Pinglay, principal investigator with AI BioDesign
view moreCredit: Allen Institute
Seattle—September 3, 2026—"Endless forms most beautiful” was how Charles Darwin described the spectacular diversity of nature from which our current understanding of life emerged. But that diversity, which evolved over billions of years, is only a fraction of what could exist given the trillions of DNA sequences available in nature, suggesting that our natural world represents only a small slice of what could have been.
Today, AI BioDesign—a bold new collaborative accelerator that will use artificial intelligence, large-scale experiments, and open science to explore that uncharted design space—was announced. It is a collaboration between the Allen Institute, University of Washington and Fred Hutch Cancer Center.
Working together, the AI BioDesign team will generate open models, datasets, assays, and tools that scientists can use to create new biological solutions to improve human health, address environmental challenges, and spur advanced technologies. The goal is to learn and model the rules biology uses to build life, enabling the potential development of everything from new drugs to treat cancer and neurodegeneration, to enzymes that can break down plastics in the ocean, to biological computers that use vastly less power than current silicon chips.
“For the first time, the speed of AI is beginning to match the experimental power of synthetic biology,” said Nobel laureate David Baker, lead scientific director of AI BioDesign, director of the UW Medicine Institute for Protein Design, and a Howard Hughes Medical Institute investigator. “That changes the question from ‘what has nature already made?’ to ‘what else is possible, and how can we test it?’ AI BioDesign can help turn that vast unknown into models that can help us solve some of humanity’s hardest problems.”
Supported by Fund for Science and Technology (FFST), a private foundation in the Paul G. Allen philanthropic ecosystem, AI BioDesign brings together complementary strengths across Seattle’s scientific ecosystem: the Allen Institute’s experience building large-scale, open-science platforms; the University of Washington’s expertise in synthetic biology and genome science, such as the Institute for Protein Design and UW Medicine Brotman Baty Institute for Precision Medicine; and Fred Hutch’s depth in cellular systems, genomics, and translational medicine.
“What excites me about AI BioDesign is that it brings together the right people and the right institutions at the right time to advance biological design with AI in the loop,” said Rui Costa, president and CEO of the Allen Institute. “The Allen Institute was built for this kind of work: big science, team science, and open science that creates resources entire fields can use. AI BioDesign combines this approach with AI models to guide which data we generate next, so experiments and models improve together in a continuous cycle of learning and testing. Ultimately, that can help us design new biological functions with greater precision.”
“AI BioDesign is exactly the kind of ambitious, collaborative science FFST was created to support,” said Marc Malandro, chief programs officer and co-lead at Fund for Science and Technology. “As a foundation, we’re looking for projects and to create environments of not just a single discovery but for multiple discoveries and platforms of knowledge that we can share openly.”
How AI BioDesign will work: designing new building blocks of life
AI BioDesign will create a continuous learning platform for biological design. AI models will propose new biological designs, scientists will build and test those designs at scale, and the results will feed back into the models such that each round becomes more accurate and informative. Over time, this design-build-measure-learn cycle will help researchers move from trial and error toward more predictable biological engineering.
“Over the last two centuries, engineering has transformed the world at least three times: the Industrial Revolution, electrification and mechanization, and the digital revolution,” said Jay Shendure, lead scientific director of AI BioDesign, scientific director of the UW Medicine Brotman Baty Institute for Precision Medicine, scientific director of the Seattle Hub for Synthetic Biology, and a Howard Hughes Medical Institute investigator. “We believe engineering’s fourth act lies at the intersection of AI and biology. Biology is code that builds: DNA carries digital instructions, and cells turn those instructions into physical systems with extraordinary precision. AI BioDesign gives us a way to learn that instruction set more systematically, test it at scale, and begin designing new biological functions.”
A distinct yet complementary approach to AI-powered biology
AI BioDesign joins a fast-moving field. Around the world, teams are building foundation models, virtual-cell systems, lab-in-the-loop platforms, and AI-enabled discovery pipelines. AI BioDesign is designed to complement those efforts by creating an open, experimentally grounded research accelerator that produces reusable resources for the broader scientific community.
Its distinction lies in the combination of multiple modular models built from tractable biological problems; new data generated from designed biological sequences and perturbations; multiplex experiments that test many designs at once; and the open sharing of models, datasets, assays, reagents, and benchmarks.
“For me, biology is ultimately a design challenge,” said Sanjay Srivatsan, an AI BioDesign principal investigator and assistant professor at the Fred Hutch Cancer Center. “As part of AI BioDesign, our team plans to vastly scale up the number of genomic datasets available to researchers. We can then use AI to understand biological patterns in those datasets and use those patterns to inspire solutions to biological problems, such as designing cells that can remove cancer from the body.”
About Allen Institute
Allen Institute is a 501(c)(3) nonprofit medical research organization dedicated to accelerating science for a healthier world. Through large-scale, multidisciplinary research initiatives, the Institute generates foundational knowledge, data, tools, and models that are shared openly with the world to advance our understanding of life and health. Founded by Jody Allen and the late Paul G. Allen, Allen Institute is supported primarily by Fund for Science and Technology.
About the University of Washington
The University of Washington is one of the world's leading public research universities with campuses in Seattle, Bothell, and Tacoma, as well as a world-class academic health system integrating research, education, and clinical care serving Washington state and the Pacific Northwest. The UW Medicine Institute for Protein Design, led by Nobel laureate David Baker, builds new, custom-designed proteins using advanced software and AI to solve major global challenges in medicine, technology, and sustainability. The Brotman Baty Institute for Precision Medicine (BBI) is a collaborative research hub based at UW Medicine that was launched in 2017 to advance the field of precision medicine. Funded by a gift from Jeff and Susan Brotman and Dan and Pam Baty, the institute combines the resources and scientific expertise of three top-tier institutions: UW Medicine, the Fred Hutch Cancer Center, and Seattle Children's. Led by Dr. Jay Shendure, internationally renowned professor of genome science at University of Washington’s School of Medicine, BBI focuses on developing and deploying cutting-edge technology in genomics, single-cell analysis, and translational research to accelerate scientific discoveries and directly improve patient care across these premier organizations.
About Fund for Science and Technology
Launched in 2025, Fund for Science and Technology is a 501(c)(3) private foundation and part of the Paul G. Allen philanthropic ecosystem. It funds new ways for scientists and researchers to collaborate and innovate in three focus areas, including bioscience, environment, and AI-for-good. Through his estate, Paul G. Allen (d. 2018), best known as the co-founder of Microsoft and among the world’s most generous philanthropists, directed the foundation’s formation and focus areas and provided its funding. Learn more online at www.ff-st.org.
— 30 —
New AI data center at KIT
More computing power built fast for research, digital autonomy and climate protection
image:
High‑performance interconnects in the new modular data center enable seamless networking across numerous systems. (Photo: Andreas Drollinger)
view moreCredit: Andreas Drollinger
Research in medicine, environmentally friendly energy supply and modern AI applications requires enormous computing power. At the same time, the demands placed on the energy-efficient operation of data centers are increasing. By building a new data center, KIT is now paving the way to meet the growing demand for computing power whilst making its digital infrastructure more sustainable.
“For us, digitalization and sustainability go hand in hand,” says Professor Kora Kristof, Vice President Digitalization and Sustainability at KIT. “With the new data center, we’re laying the foundation for tomorrow’s research and AI applications. At the same time, we use the waste heat generated to heat buildings on campus and optimize the data center's operations in line with sustainability criteria. This strengthens our digital infrastructure and drives KIT’s decarbonization and other sustainability efforts forward.”
Infrastructure Grows with Demand
“High-performance data centers are essential for many scientific breakthroughs and for the successful use of artificial intelligence,” says Professor Martin Frank, Director of the Scientific Computing Center (SCC) at KIT. “Whether it’s developing new medicines, understanding climate change or designing sustainable energy systems – many tough challenges simply can’t be tackled without serious computing power.”
The plant is initially expected to have an electrical output of around two megawatts. This computing power makes it possible to run several large-scale scientific simulations and AI applications at the same time. Thanks to a modular design and a power supply of up to ten megawatts, the computing capacity can be expanded as demand increases.
More Utilization, Less Downtime: AI by Day, Simulations by Night
“Computing power is a valuable resource. Our aim is to build a system that can ‘breathe’ with demand,” says Dr. Martin Nußbaumer, who also heads the SCC at KIT. “During the day, it provides more capacity for AI applications when demand for them is particularly high. At night, it uses the same infrastructure for resource-intensive simulations, for example in climate research. This helps us prevent idle capacity while also strengthening digital sovereignty through a well‑developed set of AI services that make smart use of existing resources.”
The data center is being built in the immediate vicinity of the combined heat and power plant on KIT’s Campus North. The heat generated by the computers is to be fed into the campus heating network via heat pumps. This will enable the data center to cover a large proportion of the buildings’ heating requirements in future and reduce its reliance on fossil fuels.
Construction is due to begin as soon as possible and be completed by the end of 2029. The first computers are expected to become operational in 2030. They will then replace existing systems at KIT and create additional capacity for data-intensive research.
Costing an estimated 24 million euros, the project forms part of the federal government’s fast‑track construction initiative to accelerate key infrastructure for science, digitalization and innovation.
In close partnership with society, KIT develops solutions for urgent challenges – from climate change, energy transition and sustainable use of natural resources to artificial intelligence, sovereignty and an aging population. As The University in the Helmholtz Association, KIT unites scientific excellence from insight to application-driven research under one roof – and is thus in a unique position to drive this transformation. As a University of Excellence, KIT offers its more than 10,000 employees and 23,000 students outstanding opportunities to shape a sustainable and resilient future. KIT – Science for Impact.
ERC Starting Grant: Michael Hahn to tackle fundamental flaws in large language models
image:
Computational linguist Michael Hahn has set out to strengthen the theoretical foundations of AI reasoning, and he has now been awarded a European Research Council (ERC) Starting Grant worth €1.5 million over five years to pursue this work.
view moreCredit: Thorsten Mohr/UdS
Large language models (LLMs) such as ChatGPT have become an integral part of everyday life for many people and businesses. They are being used, for example, to gather information from multiple sources and to support complex planning processes. Yet these AI systems are not infallible and they continue to make mistakes – LLMs hallucinate and struggle to combine information in a logically rigorous manner. Computational linguist Michael Hahn has set out to strengthen the theoretical foundations of AI reasoning, and he has now been awarded a European Research Council (ERC) Starting Grant worth €1.5 million over five years to pursue this work.
Large language models are being used to handle increasingly complex tasks to which they are not yet particularly well suited. Problems arise when companies use these models to manage complex planning scenarios such as coordinating staff rotas with material deliveries, production processes and delivery deadlines. ‘Language models still struggle with sequences of interdependent events, especially when the relevant information comes from different sources. When drawing up a staffing rota, for example, a model must take account of when materials will arrive, which production steps depend on them and whether enough staff are available to meet the delivery deadline. We want to understand how the theoretical foundations of language models need to be improved so that they can keep track of such interdependent sequences more reliably,’ says Michael Hahn, Professor of Computational Linguistics at Saarland University. Using large datasets, he and his team want to investigate which relationships AI systems can infer successfully from the data and identify those cases where they still fall short.
‘Large language models find it particularly difficult to distinguish between very similar pieces of information. This is the case, for example, in biomedical research, where descriptions of molecules and cellular processes may differ only in subtle respects,’ explains Michael Hahn. Hahn wants to develop a deeper scientific understanding of how AI systems learn and how they arrive at conclusions step by step. ‘Inductive bias plays a key role,’ says Hahn. ‘This is the set of assumptions and prior knowledge that a machine-learning algorithm needs in order to be able to generalize from familiar training data to data it has not encountered before.’
He and his team will examine the underlying mathematical models in depth and plan to develop a theory to explain how different training conditions affect the ability of an AI system to develop the low-level and high-level reasoning required to reach a well-founded conclusion. ‘We want to develop the theory using real-world tasks so that we can both predict and prevent model errors. Our goal is to make language models more reliable and cost-effective and, ultimately, help pave the way for trustworthy AI,’ says Hahn.
Michael Hahn has now been awarded an ERC Starting Grant worth €1.5 million over five years for his project ‘REALM: Foundations for Reliable Language Model Reasoning via Inductive Biases’. Last year, he received a similar amount through the German Research Foundation’s Emmy Noether Programme to gain a better understanding of the fundamental architecture of language models and explore new approaches to their design (see press release of 19 November 2025). In March, Michael Hahn was also awarded the Heinz Maier-Leibnitz Prize.
Computational linguist Michael Hahn is a member of Saarland University’s Department of Language Science and Technology, which conducts internationally renowned research and teaching at the intersection of language, cognition and artificial intelligence. The department maintains close links with the Saarland Informatics Campus. In the latest funding round, Yiting Xia and Thomas Leimkühler of the Max Planck Institute for Informatics also received ERC Starting Grants. This brings the number of researchers at the Saarland Informatics Campus who have secured one of the European Union’s various ERC grants to 54. Of the 421 ERC Starting Grants awarded in the current round, 22 are in the field of computer science. Only five of these went to Germany, three of them to the researchers at the Saarland Informatics Campus named above.
Background information – Saarland Informatics Campus
One thousand scientists and about 2,800 students from more than 81 nations make the Saarland Informatics Campus (SIC) one of the leading locations for computer science in Germany and Europe. Four world-renowned research institutes, namely the German Research Center for Artificial Intelligence (DFKI), the Max Planck Institute for Informatics, the Max Planck Institute for Software Systems, the Center for Bioinformatics as well as Saarland University with three departments and 24 degree programs cover the entire spectrum of computer science. https://saarland-informatics-campus.de/
No comments:
Post a Comment