HieraScaffold is an AI framework that could help autonomous-driving and robotics teams create large, dynamic simulation environments with less reliance on costly real-world LiDAR data collection.

Finding better materials to store hydrogen safely and efficiently is key to clean energy, but AI models trained on incomplete data can recommend candidates that don't actually work in practice. An international team led by Tohoku University, drawing on insight from 69 researchers, has proposed a "physics-aware" framework that combines thermodynamic constraints, reproducible data, and real experimental feedback to make AI-guided materials discovery more reliable. The roadmap could help accelerate the search for practical hydrogen storage solutions to support future clean-energy systems.

Stowers Institute for Medical Research Investigator Julia Zeitlinger, Ph.D., and her team partnered with Google DeepMind on AlphaGenome Atlas, a new AI-generated resource released today that predicts the molecular effects of more than 9 billion possible single-letter DNA changes across the human genome. Zeitlinger's lab used the Atlas to map regulatory motifs — the short DNA sequences that help determine which genes are turned on and off — at base-pair resolution across hundreds of cell types, work that would be impractical to do experimentally. The one-petabyte resource is searchable through a web browser without writing code, allowing researchers to prioritize which genetic variants and biological mechanisms to investigate first. Researchers from Google DeepMind, Stowers, Broad Institute, the University of Exeter, Memorial Sloan Kettering Cancer Center and Stanford University contributed scientific input and tested applications of the resource.

Published chemical literature has accumulated a wealth of reaction data, but this data comes from diverse sources, is unevenly distributed, and is clearly biased towards highly selective "successful experiments." Can such literature data be used for reliable machine learning modeling and further guide new catalytic reactions? Professor Luo Sanzhong's team and collaborators at Tsinghua University recently reported OrgAIcat, a machine learning framework for asymmetric organocatalysis. The research team manually compiled over 22,000 data points on Aldol and Michael addition reactions from 844 papers to construct the iSynth dataset. Based on this, the team developed R-SPOC, a reaction descriptor that requires no quantum chemical calculations, for predicting enantioselectivity and screening catalysts. When OrgAIcat was combined with Bayesian optimization, a challenging Aldol reaction was improved from 41% ee to 96% ee, with the entire optimization process requiring only 12 experiments. These results demonstrate that after rigorous compilation, proper characterization, and sufficient validation, historical literature data can still be used to guide experiments, despite its heterogeneity and publication bias. The article was published as an open access Research Article in CCS Chemistry, the flagship journal of the Chinese Chemical Society.

Excess nitrogen entering rivers, lakes, and groundwater is a major environmental challenge worldwide. Fertilizers, livestock waste, domestic sewage, atmospheric deposition, and other diffuse sources can all contribute to nitrogen pollution, yet determining exactly where the nitrogen comes from and how it moves through a watershed remains difficult.

Northwestern Medicine scientists have developed an AI tool that rapidly and successfully measures colonoscopy quality by reviewing footage of the procedure.

In a new paper published in Nature Communications, Michele Simoncelli, assistant professor of applied physics at Columbia, sets a benchmark for evaluating ML models that aim to predict the thermal and mechanical properties of different materials. The benchmark, developed with colleagues Balázs Póta, Paramvir Ahlawat, and Gábor Csányi at the University of Cambridge, adds a critical new level of physics-awareness to computer-driven outputs.