The day artificial intelligence took over academia

Agents4Science flipped academic norms by requiring all 48 accepted papers to list an AI as lead author and undergo AI-led peer review. Organizers tested whether large language models can generate hypotheses, run analyses, and write manuscripts; GPT-5, Gemini 2.5 Pro, and Claude Sonnet 4 scored submissions before human review. Results were mixed: AI did most hypothesis generation in just over half of papers, and helped with writing in 90%, yet often misread methods, produced buggy code, and fabricated citations. A highlighted study used ChatGPT models to design four-helix-bundle proteins, yielding one lab-validated success amid many low-confidence sequences. Critics argue scholarship requires human judgment and “taste,” which AI lacks; supporters say transparent trials can shape responsible guidelines and ease reviewer burdens in overstretched publishing systems today. https://www.science.org/content/article/futuristic-meeting-ais-took-lead-producing-and-reviewing-all-studies

Alzheimer’s disease disrupts brain-cell circadian rhythms

Researchers at Washington University School of Medicine used mouse models to show that in Alzheimer’s-type brains the usual 24-hour gene-expression rhythms in microglia and astrocytes (cells that clear amyloid plaques) become severely disrupted. They found amyloid accumulation reprograms the daily timing of hundreds of genes, including about half of the known Alzheimer’s-risk genes. The disrupted rhythms impair the brain’s waste-clearing functions and may help explain early sleep/wake impairments in patients. The authors suggest restoring these circadian rhythms in support cells might offer a new therapeutic angle for Alzheimer’s disease. (Washington University News)

Brain-inspired computing device aims to slash AI energy use

At Massachusetts Institute of Technology, PhD student Miranda Schwacke is developing ionic-synapse devices for neuromorphic (brain-inspired) computing. Unlike conventional computers where storage and processing are separate, her device combines both in one medium, mimicking neuronal/synaptic behaviour. The target is to drastically reduce the energy cost of training and running large AI-models by borrowing from how brains learn. She studies oxide thin films whose electrical conductivity can be tuned like synaptic strength, aiming toward energy-efficient hardware for the next generation of AI. This direction suggests a convergence of neuroscience, materials science and computing architecture. (MIT)

Engineered vitamin K analogues promote neuron regeneration

Scientists at Japan’s Shibaura Institute of Technology have developed enhanced analogues of vitamin K (combined with retinoic acid) that cross the blood-brain barrier and strongly stimulate neural progenitor cells to differentiate into neurons. These new compounds were about three-times more potent than natural vitamin K in driving neuron formation, via the mGluR1 receptor. Because neuron loss drives diseases such as Alzheimer’s and Parkinson’s, the authors suggest these analogues may be starting points for regenerative therapies. While still early, the work opens a path to replenishing lost neurons rather than only slowing degeneration. (Science Daily)

Tiny brain nanotube channels may spread Alzheimer’s pathology

A team at Johns Hopkins University discovered that neurons form microscopic tunnelling nanotubes (TNTs) through which they transfer materials — normally waste-clearing—but in Alzheimer’s models these channels seem to propagate toxic proteins like amyloid-β between neurons. In mice with Alzheimer’s-type pathology, these nanotubes increased in number and may serve as a conduit for spread of neuropathology. The finding suggests that targeting these inter-cellular nanotube networks could offer a novel therapeutic angle to slow disease spread rather than just tackling plaques in isolation. (Science Daily)

AI-powered map reveals uncharted regions of the mouse brain

Researchers at University of California, San Francisco and the Allen Institute for Cell Science used a ChatGPT-style model to generate a detailed brain map of the mouse including roughly 1,300 distinct regions without manual annotation. The AI model traced cells and defined boundaries using high-throughput imaging data. By automating the atlas-building process, the method promises to accelerate the mapping of other organs and species, including humans. The automated map lays a stronger foundation for linking cell-type, location and function in brain circuits and could hasten discovery of brain-disease mechanisms. (Live Science)

---

---

Rare MC4R gene mutation links obesity with lower heart-disease risk

A genetic-epidemiology study found that individuals with rare disabling mutations in the appetite-regulating gene MC4R – who are typically obese – appear to have lower LDL cholesterol, lower triglycerides, lower blood pressure and about half the risk of coronary artery disease compared to equally obese people without the mutation. The finding challenges the simple “obesity = cardiovascular risk” narrative and points to the idea that MC4R has direct metabolic effects on fat and lipid handling, beyond appetite control. The insight may guide development of drugs that mimic the heart-healthy effects of MC4R deficiency without causing large weight gains. (Medical Express)

Brain region tied to chronic pain may enable new therapies

In a mouse-study published recently, scientists have identified a specific neuronal ensemble in a previously overlooked brain region that becomes highly active and persistent after injury, potentially serving as a “chronic pain circuit”. The region appears to enable the transition from acute to persistent pain. The discovery offers the possibility of therapies that interrupt this circuit and prevent chronic pain from taking hold — a major unmet clinical need, given the prevalence of long-term pain. (Nature)

Ancient lead exposure may have shaped human brain evolution

An international team analysed 51 fossil hominid and ape teeth, discovering chemical “lead bands” that indicate intermittent lead exposure over two million years, long before modern industrial pollution. They combined the data with brain-organoid experiments carrying Neanderthal and modern human genes, and found archaic-gene organoids were more vulnerable to lead-induced disruption of FOXP2- and NOVA1-related circuits (relevant to speech and cognition). The authors suggest that environmental lead may have influenced brain-development and behavioural evolution in our ancestors. This adds a toxic-metal dimension to the story of how human cognition diverged from our hominid relatives. (Phys.org)

Integrative roadmap outlines future of neurotechnology

A new perspective paper outlines a strategic roadmap for the development of future neurotechnologies — devices that sense, compute and stimulate the brain and body in real-time. The authors (mainly early-career scientists) highlight five major trade-offs (functionality vs scalability, adaptability vs reliability, etc.) and stress the need for interdisciplinary collaboration across neuroscience, materials science, signal-processing, AI and ethics/regulation. They call for standardised platforms, shared datasets and clear pathways to translation and societal impact. The paper may shape how neurotech moves from lab-prototype to safe, equitable clinical and consumer use. (Arxiv)

Sustainable AI via brain-inspired computing device

A materials-science/engineering team at MIT has created an electrochemical device whose behaviour mimics synaptic plasticity. In this “ionic synapse” platform, the same material region stores data and performs computation. Because conventional digital computers separate storage and processing, they waste significant energy moving data back and forth; by contrast the brain-inspired architecture uses orders-magnitude less energy. The researchers propose this kind of hardware for future AI systems where energy cost is a bottleneck. Their work bridges neuroscience (synaptic mechanisms) with next-generation computing hardware. (MIT)

Did Tides, Not Canals, Seed Sumerian Cities?

New research in PLOS ONE argues early Mesopotamian farming exploited twice-daily tides, not state-run canal megaprojects. A 25-meter soil core from Lagash, spanning ~7000 years, revealed marine shells and coastal sediments, placing the site near a tide-influenced delta. The team proposes denser saltwater pushed upriver beneath freshwater, lifting river levels on a reliable schedule villagers diverted to irrigate wheat, dates, and gardens long before cities. That steady water and fresh silt enabled population growth and occupational diversity; when tidal predictability waned, communities turned to engineered canals around 2500 B.C.E. Supporters say this recasts Sumerian urbanism as bottom-up. Critics warn about brackish-water complications, the modest scale of early irrigation, abundant wetland resources that could sustain towns, and the risk of generalizing a civilization-wide story from a single core. (Science)