Solving The Money Problem
Get the latest international news and world events from around the world.
How scientists got a glimpse of the inner workings of protein language models
Now, a team of researchers based at the Massachusetts Institute of Technology (the United States) has tried to shed light on the inner workings of the language models that predict the structure and function of proteins by using an innovative technique. They have described their findings in the study, ‘Sparse autoencoders uncover biologically interpretable features in protein language model representations’, which was published in the journal Proceedings of the National Academy of Sciences last month. The team included Onkar Gujral, Mihir Bafna, Eric Alm, and Bonnie Berger.
Story continues below this ad.
Berger, the senior author of the study, told The Indian Express over email, “This is the first work that allows us to look inside the ‘black box’ of protein language models to gain insights into why they function as they do.”
Scientists Turned Our Cells Into Quantum Computers—Sort Of
For the protein qubit to “encode” more information about what is going on inside a cell, the fluorescent protein needs to be genetically engineered to match the protein scientists want to observe in a given cell. The glowing protein is then attached to the target protein and zapped with a laser so it reaches a state of superposition, turning it into a nano-probe that picks up what is happening in the cell. From there, scientists can infer how a certain biological process happens, what the beginnings of a genetic disease look like, or how cells respond to certain treatments.
And eventually, this kind of sensing could be used in non-biological applications as well.
“Directed evolution on our EYFP qubit could be used to optimize its optical and spin properties and even reveal unexpected insights into qubit physics,” the researchers said. “Protein-based qubits are positioned to take advantage of techniques from both quantum information sciences and bioengineering, with potentially transformative possibilities in both fields.”
How early brain structure primes itself to learn efficiently
Vision happens when patterns of light entering the eye are converted into reliable patterns of brain activity. This reliability allows the brain to recognize the same object each time it is seen. Our brains, however, are not born with this ability; instead, we develop it through visual experience. Collaborating scientists at MPFI and the Frankfurt Institute for Advanced Studies have recently discovered key circuit changes that lead to the maturation of reliable brain activity patterns.
Their findings, published in Neuron this week, are likely generalizable beyond vision, providing a framework to understand the brain’s unique ability to adapt and learn quickly during the earliest stages of development.
The brain is a highly organized structure. Like other brain regions, visual areas have structure to them, which scientists call modules. This modular organization consists of patches of neurons that activate together in response to specific information. For example, some patches of neurons activate together in response to seeing vertical stripes, while other patches activate when horizontal stripes are seen.
Alleviating head-mounted weight burden for neural imaging in freely-behaving rodents
Liu et al. present a remarkably simple yet clever method of mitigating the effects of head-mounted microscopes on mouse behavior: they tethered a helium balloon to the microscope device to counter its weight! A fun and useful engineering solution!
Link to article.
Scientific Reports — Alleviating head-mounted weight burden for neural imaging in freely-behaving rodents. Sci Rep 15, 19175 (2025). https://doi.org/10.1038/s41598-025-04300-0
An American Collider Is Finally Ready to Recreate Matter from the Beginning of Time
Today, the absolute heart of particle physics is located in Geneva, Switzerland at CERN’s Large Hadron Collider. This instrument’s unmatched size, power, and precision make it the ultimate tool for exploring high-energy particle physics. However, one tool can’t do everything, and even immensely useful ones like the LHC sometimes need a helping hand.
That’s where Brookhaven National Laboratory’s (BNL) Relativistic Heavy Ion Collider (RHIC) comes in. In 2015, the U.S. Department of Energy approved an upgrade to the Pioneering High Energy Nuclear Interaction eXperiment (PHENIX)—an instrument originally designed to explore the components of the quark-gluon plasma (QGP) that formed one millionth of a second after the Big Bang. According to Edward O’Brien (a physicist from BNL), the idea behind this super PHENIX, or sPHENIX, was to “provide physics results which focused on jets and heavy flavor [of quarks] that complemented and overlapped with the Heavy Ion physics results being generated by the experiments at the CERN Large Hadron Collider.”