A new AI tool developed by Mass General Brigham has identified long COVID in 22.8% of patients, much higher than previous 7% estimates. The tool analyses health records from nearly 300,000 patients, using 'precision phenotyping' to distinguish long COVID symptoms from pre-existing conditions. It improves diagnostic accuracy and addresses demographic biases. Researchers plan to release the algorithm publicly for global use, potentially advancing research on long COVID subtypes.
A major AI-ready dataset on type 2 diabetes has been released, containing diverse data from over 1,000 participants. The AI-READI study aims to collect data from 4,000 people for global analysis. The dataset is accessible through an online platform and has already been downloaded by numerous research organisations worldwide. Multiple institutions are involved in this project, based in Seattle.
AI and machine learning are revolutionising scientific prediction across fields like college admissions, elections, and drug discovery. Large datasets and complex 'black box' models are being used, with statisticians developing techniques to quantify uncertainty without understanding the models' inner workings. Concerns about reproducibility exist, and statisticians are creating safeguards for reliable findings. The emerging field of data science incorporates traditional statistics with new techniques like large-scale population tracking.
AI is rapidly advancing in simulating physics and chemistry problems, potentially rivalling quantum computers. Neural networks can now model systems of up to 100,000 atoms, making previously difficult problems in chemistry and materials science feasible. This progress is driven by improved techniques and vast datasets. While quantum computers may still have advantages for certain problems, AI could reach important milestones in chemistry and materials science simulation sooner.
ChatGPT 4 Omni, an AI system, accurately identified pills in a controlled test, showing potential for EMS training in overdose scenarios. However, its real-world application faces limitations and privacy concerns. EMS professionals are advised to understand the technology's capabilities and limitations before considering integration into practice.
Argonne National Laboratory researchers have developed MProt-DPO, an AI-driven computing framework for accelerating protein design. It combines various data sources and uses large language models and supercomputers to process billions of parameters, achieving over one exaflop of performance. The framework incorporates a Direct Preference Optimization algorithm to improve design reliability and is being tested on protein design challenges, with plans for laboratory validation.
Stanford Medicine researchers developed an AI method to identify complex structural variants in the human genome with 95% accuracy. The study analysed over 4,000 genomes worldwide, focusing on schizophrenia and bipolar disorder. The AI identified over 8,000 distinct variants, many in regions regulating brain development. This could improve understanding of heritable psychiatric conditions and potentially lead to more precise treatment approaches.
