This week in toast, AI designs cosmic detectors humans can't understand, robots learn from YouTube, and for some reason, tardigrades get inked with university logos. Plus, how blocking a single molecule could battle everything from cancer to baldness

An AI algorithm named 'Urania' now designs gravitational wave detectors far better than humans can. After two years of development, the AI has discovered dozens of new detector designs that outperform human experts' blueprints. Many of these solutions are so unorthodox that researchers have yet to fully understand how they work. 

The detectors themselves capture ripples in spacetime caused by extreme events like colliding black holes, a phenomenon Einstein predicted over a century ago but that wasn't directly detected until 2016. Designing these ultra precise instruments is fiendishly difficult, which is likely why scientists turned to AI for help.

🧐 What's in it for me? Improved gravitational wave detectors could expand our ability to detect signals from deep space by more than an order of magnitude. This translates to a vastly expanded cosmic eavesdropping network, allowing us to observe more extreme events across the universe. If the universe is having conversations, we've just upgraded from a tin can telephone to fibre optics.

💵 Out of the Lab: Although early for commercialisation, research institutions like LIGO Scientific Collaboration are exploring ways to implement these AI-designed detectors for the next generation of gravitational wave observatories. The AI approach used by Urania could also influence how we design other scientific instruments and technologies. As researcher Mario Krenn puts it, "We are in an era where machines can discover new super human solutions in science, and the task of humans is to understand what the machine has done." Which is a polite way of saying "the AI is smarter than us and we're trying to catch up."

A new framework called RHyME enables robots to learn tasks by watching just a single how-to video. Previous robot training required tedious, step by step instructions and would collapse if anything went off script, like dropping a tool. With RHyME, robots can now use their "memory" to connect the dots when performing tasks they've only seen once.

This is akin to how Tesla trained their autonomous vehicles, which proved so much more efficient than trying to hard code road rules. The system requires just 30 minutes of robot data and achieved a more than 50% increase in task success compared to previous methods.

🧐 What's in it for me? This technology could significantly reduce the time, energy, and money needed to train robots, making them more accessible and versatile. Future home robot assistants may become viable much sooner as they gain the ability to adapt to real world environments and the countless contingencies they present. 

💵 Out of the Lab: Companies like Everyday Robotics and Figure AI are racing to develop adaptable home assistant robots that could benefit enormously from this type of one shot learning. Boston Dynamics, known for their agile robots, could integrate this technology to make their machines not just physically capable but also quicker to train for new tasks. The reduced data requirements also mean smaller companies could enter the robotics market with less initial investment, potentially disrupting incumbent players.

The RoboBee has received a crucial upgrade: landing gear. Inspired by the crane fly, the team equipped their insect sized robot with long, jointed legs and an improved controller that helps it decelerate on approach for a gentle touchdown. Previously, scientists would have to turn off the robot above the ground and just drop it, "praying" it would land upright without damaging its delicate components. The RoboBee weighs just a tenth of a gram with a wingspan of 3 centimetres, making precise landings particularly challenging. 

🧐 What's in it for me? The RoboBee's upgrades are a critical step toward fully autonomous micro robots that could eventually be used for environmental monitoring, disaster surveillance, and even artificial pollination. While it currently remains tethered to off board control systems, the Harvard team continues working toward scaling up the vehicle and incorporating onboard electronics for complete autonomy. Imagine swarms of RoboBees buzzing around vertical farms and gardens of the future, replacing the BioBees we've been so cleverly killing off.

💵 Out of the Lab: While the RoboBee isn't yet commercialised, several startups are exploring robotics for various Agri applications. Dogtooth Technologies is developing autonomous berry-picking robots that require similar delicate precision and Muddy Machines is creating agricultural robots for vegetable harvesting. As habitat loss and pesticides continue to threaten natural pollinators, the artificial pollination market alone could become a multibillion dollar opportunity within the decade.

When metals and ceramics are exposed to heat (like in airplane or rocket engines), their internal structures of crystals or "grains" can change. Some grains grow abnormally large, which can transform a flexible material into something brittle. Until now, finding stable materials has been like searching for a needle in a haystack, requiring expensive and time consuming testing of countless combinations.

Now there is a new machine learning model that can predict abnormal grain growth in simulated polycrystalline materials (computer models that replicate metals and ceramics made of multiple tiny crystals), a common cause of material failure under continuous heat. The model accurately predicts 86% of abnormal grain growth cases within the first 20% of the material's lifetime, allowing engineers to identify potentially unstable materials far in advance of actual failure. Think of it like a psychic for your rocket engine, but with actual science behind it.

🧐 What's in it for me? This breakthrough could lead to stronger, more reliable materials for high stress environments like combustion engines, allowing them to run hotter and longer before failing. For the aviation and aerospace industries, this means safer, more efficient engines and potentially significant cost savings. For the rest of us, it means planes that don't fall out of the sky and rockets that actually make it to orbit, which seems rather important.

💵 Out of the Lab: Companies like General Electric Aviation and Pratt & Whitney could implement this technology to accelerate their development of next generation engine materials. SpaceX and Blue Origin might leverage similar techniques to design more reliable rocket components capable of withstanding extreme conditions. The researchers also believe their method could predict other rare events beyond materials science, potentially helping to forecast phase changes in materials, dangerous pathogen mutations, or sudden atmospheric shifts.

Scientists have finally worked out the structure and function of the mitochondrial pyruvate carrier, a molecular machine that transports pyruvate (a molecule generated from sugar breakdown) into our mitochondria, the powerhouses of our cells. The carrier works like a canal lock, with an outer gate opening to let pyruvate in, then closing while an inner gate opens to allow passage into the mitochondrion.

First proposed to exist in 1971, scientists have now visualised this structure at the atomic scale using cryo electron microscopy. The breakthrough could lead to new treatments for a range of conditions by targeting this crucial cellular gateway.

🧐 What's in it for me? Because of its central role in controlling how mitochondria produce energy, this carrier is now recognised as a promising drug target for conditions including diabetes, fatty liver disease, Parkinson's disease, and even hair loss. Blocking the pyruvate carrier would force the body to look elsewhere for fuel, potentially treating conditions like fatty liver disease by encouraging the body to use fat stored in liver cells. For specific cancers that rely heavily on pyruvate metabolism, such as some types of prostate cancer, blocking the carrier could starve the cancer cells of the energy they need to survive. 

💵 Out of the Lab: Pharmaceutical companies like Metabolic Solutions are already exploring compounds that target mitochondrial metabolism for various diseases. The hair regrowth industry, worth over £7 billion globally, will likely take particular interest in this research, with companies like Aurobindo Pharma who may want to explore how pyruvate carrier inhibition might be incorporated into next generation hair loss treatments. Cambridge Enterprise, the University's commercialisation arm, has reportedly filed patents on several molecules that target this carrier with remarkable specificity.

A deep learning algorithm can now identify potential disease causing mutations in the noncoding regions of DNA, which make up over 98% of the human genome. These regions don't code for proteins (which are a lot of what we’re made of) but often control when proteins are made. The algorithm successfully detects "footprints" where proteins called transcription factors bind to regulate gene expression, helping pinpoint which specific variants among several might be responsible for disease.

Previous studies identified broad regions associated with disease, but pinpointing the exact culprit variant remained challenging. As one researcher put it, "This situation is comparable to a police lineup. You're looking at similar suspects together, so it can be challenging to know who the actual culprit is." With this new approach, they can finally identify the genomic perp with confidence.

🧐 What's in it for me? This technique could ultimately inform the design of novel treatments for common diseases by helping scientists understand the specific genetic variants driving them. By resolving fundamental issues in identifying disease causing noncoding variants, researchers can better target their therapeutic approaches. For patients with genetic diseases, this could lead to more precise treatments tailored to their specific genetic profile.

💵 Out of the Lab: Genomics companies like Illumina and Oxford Nanopore Technologies could incorporate these techniques into their analysis platforms to provide more comprehensive health insights. Pharmaceutical giants including Regeneron and Novartis are investing heavily in genetic analyses to inform drug discovery, and this method could become an essential part of their toolkit, whilst startups like Variant Bio and Deep Genomics, which use AI to analyse genetic data for therapeutic development, are particularly well positioned to implement these techniques.

New research has discovered that interactions between immune and brain cells drive fear responses, and treatment with psychedelics like MDMA and psilocybin may reverse these effects. The study found that increased crosstalk between cells in the amygdala (the brain's fear centre) boosts fear behaviours, elevates inflammatory signalling, and activates fear promoting neurons. During chronic stress, inflammatory immune cells called monocytes migrate to the brain meninges (protective layers of tissue that encase the brain and spinal cord), and manipulating these cells impacts fear behaviours.

When stressed mice were treated with psilocybin and MDMA, it prevented monocytes from accumulating in the brain and lowered fear behaviours. Similar signals were found in human brain cells and in gene expression datasets from patients with major depressive disorder, suggesting that these same interactions may play a role in human neuropsychiatric disorders.

And it seems Americans have taken notice. Psilocybin use has increased significantly nationwide since 2019, particularly among those with mental health conditions or chronic pain. Lifetime use among adults rose from 10% in 2019 (about 25 million people) to 12.1% in 2023 (over 31 million people), while past-year use increased by 44% among young adults and a whopping 188% among adults over 30. More adults now use psilocybin than drugs like cocaine, LSD, methamphetamine, or illegal opioids. The timing aligns with some US states decriminalising or legalising psilocybin. 

🧐 What's in it for me? This research suggests that psychedelics could help treat not only mental health conditions but also inflammatory diseases through their effects on both brain cells and immune cells. It helps explain why psychedelics have shown promise in treating depression, anxiety, and PTSD. However, the researchers emphasise that "we're not saying that psychedelics are a cure all for inflammatory diseases or any other health condition," but rather that they have "tissue specific benefits" that could open up new treatment possibilities.

💵 Out of the Lab: Clinical trials of psychedelic treatments for mental health conditions are already underway at companies like COMPASS Pathways, whilst companies like Atai Life Sciences are developing psychedelic inspired therapeutics that could target these neuroimmune interactions. As more states consider regulation or legalisation, there's also growing interest in developing tools for detection and monitoring, with companies like Hound Labs exploring technologies to detect recent psilocybin use.

Scientists have found a way to tattoo Water Bears… 🧐

Microscopic tardigrades (water bears) have received tiny "tattoos" using a technique called ice lithography. These nearly indestructible creatures, known for surviving extreme conditions from freezing temperatures to outer space, served as the perfect canvas for testing a microfabrication technique for building microscopic, biocompatible devices.

About 40% of the tardigrades survived their tattoo session and showed no changes in behaviour afterwards. Writers note; if your local tattooist boasts the same survival rates, probably best avoid them. 

The scientists put the tardigrades into a suspended animation state by dehydrating them, then cooled them to -226 degrees Fahrenheit and covered them with a protective organic compound. An electron beam then drew patterns as small as 72 nanometers wide, including squares, dots, lines, and even a university logo.

The researchers believe this technique could enable printing micro electronics or sensors onto living tissue, potentially advancing biomaterial devices and biophysical sensors "that were previously only present in science fiction", and tardigrade nightmares…

Until next time, stay curious.

Like what you're reading? Share toast with a friend.