Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain
the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in
Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles
and JavaScript.
Wang, Yang et al. propose KnowDDI, a graph neural network that leverages biomedical knowledge graphs for drug-drug interaction predictions. The model yields improved performance and interpretability over existing methods, especially in scenarios with sparse knowledge graphs, marking a significant advancement in biomedicine and healthcare.
Ptacin et al. utilize a semi-synthetic microbial platform with an expanded genetic code to discover a PEGylated IL-2 compound that stimulates Tregs, with minimal effects on effector populations. Studies in mice and primates demonstrate Treg stimulation and support development of this compound for the potential treatment of autoimmune diseases.
Robitaille et al. report findings from a phase IIb randomized placebo-controlled trial evaluating the effect of a long-chain omega-3 fatty acid MAG-EPA dietary supplement on prostate cancer proliferation. The dietary supplement had no effect on the primary outcome of prostate cancer proliferation according to Ki-67 expression.
Le Buanec, Schiavon, Merandet et al. study the underlying immune responses for the elevation of IFNα that occurs during an antiretroviral treatment interruption in HIV patients. IFNα mediated induction of CCR5 is shown to promote pathogenic phenotype, while elite controllers avoid this by an unknown genetic factor or a low inoculum infection.
Le Buanec et al. compared distribution and frequency of cell markers associated with immune dysfunction between HIV elite controllers (EC) and untreated non-EC. They demonstrate that untreated HIV-infected individuals exhibit structurally and functionally impaired immune subsets, as a consequence of excessive levels of serum IFNα.
Wu, Zeng et al. assess the feasibility and efficacy of AAV9:PKP2 gene therapy in the Pkp2-cKO mouse model of arrhythmogenic right ventricular cardiomyopathy. They show that it ameliorates ventricular arrhythmias, reverses adverse right ventricular remodeling, improves heart function, and reduces mortality in this mouse model.
Kjerulff et al. evaluate 47 biomarkers, including markers of inflammation and vascular stress, and their associations with demographic and lifestyle factors in healthy participants from the Danish Blood Donor study. Circulating biomarker levels varied according to sex, age, BMI and smoking status.
Mujahid et al. develop a type 1 diabetes patient simulator using a conditional sequence-to-sequence deep generative model. Their approach captures causal relationships between insulin, carbohydrates, and blood glucose levels, producing virtual patients with similar responses to real patients in open and closed-loop insulin therapy scenarios.
Czech et al. develop and clinically validate a sensor-based approach to measure upper and lower body bradykinesia in an early Parkinson’s disease population. Results demonstrate enhanced sensitivity of sensor-based digital measurements to disease progression over one year relative to current clinical measurement standards.
Geanes et al. demonstrate that COVID-19 severity is associated with increased levels of IgG, IgA, IgM antibodies to ACE2 and other immune molecules. These findings expand the understanding of immune mechanisms in COVID-19 and suggest that the evaluation of autoantibody levels to immune factors may serve as a potential biomarker for COVID-19 severity.
Arslan et al. present the results of a comprehensive pan-cancer study evaluating deep learning-based multi-omic biomarker profiling using H&E-stained whole slide images. They show that deep learning can predict a wide range of biomarkers across the omics spectrum and in different cancers directly from histomorphology.
Tayebi Arasteh, Ziller et al. investigate how strict privacy safeguards affect AI learning in medical imaging. Their study finds that while enforcing privacy leads to a modest drop in accuracy, it does not exacerbate biases for different patient groups, though accuracy for intricate cases and specific subgroups may be more affected.
Foldyna et al. demonstrate that deep learning-based quantification of epicardial adipose tissue from low-dose chest CT scans independently predicts all-cause and cardiovascular mortality in heavy smokers. This can enhance cardiovascular risk stratification beyond traditional measures.
Bou-Nassif et al. develop a smartphone app using AI and stimulated Raman histology to differentiate pituitary adenomas from normal tissue in real-time during surgery. Prospective validation on 40 patients shows high sensitivity and specificity, with an external validation on 40 additional adenoma tumors.
Lin et al. develop an artificial intelligence model capable of using electrocardiograms (ECGs) to predict hyperthyroidism. This predictive model performs well at detecting overt hyperthyroidism and can also be used to assess the risk of mortality and heart failure, independent of laboratory based thyroid tests.
Xue et al. use 11 Mendelian Randomization methods to investigate the causal relationship between seven substance use behavioural traits and health outcomes. They identify risk effects of alcohol consumption as well as dosage dependent effects of coffee and tea on cardiovascular disease and osteoarthritis respectively.
Zhao et al. investigate optimal strategies to detect COVID-19 features in lung ultrasound images using deep neural networks trained with simulated and in vivo datasets. Including simulated data during training improves detection performance and training efficiency and is a promising alternative to curating thousands of patient images.
Cheng, Bai et al. uncover associations between 63 blood and urine biomarkers and treatment-resistant schizophrenia using data from the UK Biobank. They find correlations between some blood and urine biomarkers and treatment resistant schizophrenia polygenic risk scores.
Tsouka et al. compare metabolic signatures of metabolic dysfunction-associated steatohepatitis in a diet- and chemical-induced mouse model with human disease. By evaluating perturbations in enzymatic reactions via transcriptomics-driven metabolic pathway analysis, they observe similar alterations in lipid metabolism between the mouse model and humans.
Dempsey et al. present a graphene-based biosensor technology to detect enzyme activity in serum samples. A model is developed based on the activity of a panel of these biosensors to classify 90% of patients with lung cancer across all stages of disease, providing a potentially useful screening technology.