New papers on AI for science & medicine
10 new papers on ai for science & medicine in the last 7 days, within AI & machine learning. These are the 10 Pipette rates most worth reading, with the main result in the authors' own words.
The best of the week
All-optical sensing of dermal activity enables an AI-powered wearable ring for continuous stress monitoring
Here, we present an artificial intelligence (AI)–powered ring for all-optical sensing of dermal activity (ODA), using light rather than electrodes to track sweat gland activation, thereby enhancing durability and comfort.
Peer-reviewed journalBold claims, read criticallyReal-world useScaling medical imaging report generation with multimodal reinforcement learning
Here we show that our method improves report quality and generalization: UNIRG-CXR surpasses prior models on the ReXrank benchmark across four datasets and two generation settings, exceeding the previous best score by over 50% on two test sets, and produces fewer clinically significant errors while maintaining performance on unseen data sources and across demographic subgroups.
Peer-reviewed journalReal-world useDeep learning–assisted electron microscopy for apparent Z-height mapping of atomically dispersed catalysts
Here, we develop a physics-informed, domain-adaptive CycleGAN-UNet framework for depth-aware ADC analysis.
Peer-reviewed journalisSPA-vesicle: AI-assisted structure determination of membrane proteins on vesicles
Here, we present an AI-enabled, vesicle geometry-guided 3D reconstruction strategy based on untilted data.
Peer-reviewed journalPhysics-informed and data-efficient machine learning for storm surge forecasting during typhoons
The SSNN is iteratively trained to make rolling forecast and can produce more accurate forecasts than a standard data-driven model even with a smaller size of training data.
Peer-reviewed journalReal-world useKnowledge-guided machine learning deciphers spatiotemporal dynamics in chemical reactor systems
A physics-informed neural network embeds coking and heat transfer knowledge via an output integration layer and region-weighted loss to predict reactor spatiotemporal fields; it matches conventional simulation results, improves peak wall temperature accuracy, and runs about five orders of magnitude faster than conventional simulations.
Peer-reviewed journalBold claims, read criticallyReal-world useLeveraging sequence purification for accurate prediction of multiple conformational states with AlphaFold2
Here, we introduce AF-ClaSeq, a systematic framework that enriches and purifies state-specific evolutionary signals through iterative sequence classification and selection.
Peer-reviewed journalscMomer: a pretraining framework for inferring robust cellular representations from single-cell transcriptomes
Here, we introduce scMomer, a modality-aware pretraining framework designed for multi-modal representation learning under missing modality conditions. scMomer adopts a three-stage pretraining strategy that learns unimodal cell representations, models joint representations from multi-omics data, and distills multi-modal knowledge to enable multi-omics-like representations from unimodal input.
Peer-reviewed journalDPI-score: a deep learning-based metric for assessing protein-protein interfaces in cryoEM derived assemblies
To address this, we present DPI-Score, a deep learning-based metric for assessing protein-protein interfaces in cryoEM derived complexes.
Peer-reviewed journalImproving molecular property prediction with score-based diffusion models
Across diverse feature extractors and graph neural networks, diffusion-based training improves over conventional maximum-likelihood objectives on small-molecule regression benchmarks and matches or exceeds strong baselines on multitask regression and classification datasets.
Peer-reviewed journal