Across 11 U.S. surveys involving 183,000 adults, higher social media use was consistently associated with self-reported depression. The association in the researchers’ models shifted from neutral to positive at about 150 minutes of daily use, highlighting a question for future longitudinal and experimental studies.
A survey of snow from lowlands, mountains and polar regions found tardigrades, rotifers, nematodes, mites and springtails. The findings suggest that snowfields and small glaciers could provide refuges for some cold-loving species during warmer periods.
Researchers developed light-sensitive hydrogels that rapidly form twists, helices and other handed shapes when selected areas are illuminated. The shapes can be erased in darkness and reprogrammed with light, and the gel can rotate for robotic movement.
This review analyzed human studies comparing two treatments given before surgery for locally advanced esophageal or esophagogastric junction adenocarcinoma: FLOT chemotherapy and CROSS chemoradiotherapy. The pooled estimate favored FLOT for overall survival, but the groups had similar three-year survival, tumor response, surgical margins, and several major perioperative outcomes.
Wild relatives contain adaptive genetic diversity that was reduced during crop domestication, but conventional prediction methods often miss complex DNA changes. A review in BMC Genomics outlines how artificial intelligence and broader genetic maps could help identify these variants for climate-resilient breeding.
Researchers developed an aerial robot that uses touch to refine its position and perch on structures it has not encountered before. The system succeeded in more than 99% of simulated trials and completed 26 real-world perching trials despite corrupted position estimates.
Researchers found that information flow between major brain networks differed between autistic participants and non-autistic controls. An artificial-intelligence system classified the two groups with a mean accuracy of 91% across cross-validation, but the study does not establish a clinical biomarker.
A quasi-experimental study of 24 European Union member states found that adopting carbon taxes was associated with lower energy intensity, greater renewable energy use and higher Human Development Index scores. The gains were strongest in countries that had maintained the policies for more than two decades.
A survey of human-AI collaboration argues that more capable AI systems do not automatically produce reliable partnerships. It identifies human-centered design, governance and attention to high-stakes uses as central to making these systems trustworthy and beneficial.
Researchers used seismic signals produced when thunder’s sound waves enter the ground to image the shallow subsurface beneath an urban karst setting. The method identified several previously undetected weak zones, some of which matched surface deformation and independent underground surveys.
A study of Freedom of Information Act requests to the Securities and Exchange Commission finds that the type of requester is linked to different signals about a public company’s future stock returns. The analysis also finds evidence that institutional investors and short sellers trade on information obtained through these requests.
A review examines how pulsating stars in eclipsing binary systems can be used to study stellar interiors and the effects of tides. Space-based observations, combined with repeated measurements of starlight spectra, can provide model-independent masses and radii for thousands of systems.
Researchers observed sound waves moving ballistically through a disordered acoustic lattice even though the system’s wave states were strongly localized at irregular positions. The result separates localization of wave states from the movement of wave packets.
Deep-learning tools for predicting RNA splicing can rely on genomic clues unrelated to splicing and miss important effects of RNA structure, a study finds. These weaknesses were linked to systematic errors, especially when the tools evaluated sequences that differed from those used for training.
Researchers developed Orbformer, an AI model that learns transferable descriptions of molecules from 22,000 structures, including molecules with breaking bonds. After fine-tuning, it reached chemical accuracy on established tests and more demanding bond-breaking and Diels–Alder reaction problems, while reducing the need to solve each new molecule from scratch.
