Digitally tracking activity metrics like daily steps, hours of sleep, and average heart rates may be significantly enhanced by combining those data with biochemical measures like proteomic and genomic data, blood analyses of antibodies and metabolites, and periodic sampling of the gut microbiome, according to researchers at the University of Helsinki in Finland and Sweden’s Karolinska Institute. They followed 96 people in a 16-month health coaching program that used devices to continuously monitor their activity and sleep, and they combined those data with more than 20,000 biological samples and 53 million pieces of molecular data to give the participants personalized health coaching. Among the behavior changes noted, they report half the smokers in the study quit, almost all of them began exercising, and the prevalence of vitamin D deficiency decreased from 31 to 16 percent. Cell Systems
Researchers at the University of Copenhagen have developed a novel algorithm that could change how hospitals plan and allocate their COVID-19 resources. According to a recent study that examined data from over 42,000 Danish people, the new machine learning model can forecast with extreme accuracy for up to 10 days how many people will be admitted to the ICU and who will need a ventilator. By considering information like gender, age, BMI, smoking status, blood pressure, and more, the model provides a more detailed prediction than its competitors. Nature
How different people respond to the same viruses is remarkably varied—lately evident in the generally lower severity of COVID-19 infections in children. Looking at T cells taken from human blood samples and recovered from tissues across the body given by organ donors, scientists at Columbia University have uncovered some of the molecular and cellular factors through which things like virus type, tissue location, donor age, and sex can shape the maintenance and function of T cell immunity in the human body—which is a crucial aspect of our adaptive immune defense against viruses. They looked specifically at T cells in those samples that reflect immune responses to the common viruses influenza and cytomegalovirus, and they found both the tissues and the viruses themselves strongly influence immune memory. They also showed how age and sex play a role. Cell Reports
In the early 1970s, the “Anfinsen experiment” showed a protein’s genetically encoded sequence is what determines its 3D structure, and biologists ever since have dreamed of solving the so-called protein folding problem: To accurately predict a protein’s structure based on its sequence data alone. Last year Google DeepMind’s AI machine learning algorithm AlphaFold2 came closer than ever to achieving that, which many experts hailed as a sign of the coming revolution. This week a team at Rutgers University gave a taste of what that revolution will look like from the patient bedside, suggesting it will enable the rapid development of new drugs or vaccines based on predicted structures, like the coronavirus Delta variant spike protein, and allow personalized treatment plans for people with cancer based on predicting drug resistance due to single point mutations in their tumors. New England Journal of Medicine
The potential for artificial intelligence to improve patient outcomes is its most obvious promise for modern medicine, and in an age of big data, the crucial question is: Can a generic AI algorithm successfully predict prognoses and steer someone’s specific treatment? Focusing on the multibillion-dollar problem of unexpected hospital readmission, a team at the University of North Carolina at Greensboro showed it can. They used a machine learning algorithm trained on data collected from 76,000 people—reflecting things like demographics, medications, and comorbidities—and found they could accurately estimate the 30-day hospital readmission risk of frail patients. Patterns
Researchers at Lawrence Berkeley National Laboratory have developed a high-throughput method for identifying critical DNA sequences in bacteria known as “transcription factor binding sites,” a key part of the regulatory machinery that governs gene expression in living cells. They analyzed dozens of different bacteria and found hundreds of these binding sites, increasing the number of known transcription factor binding sites for E. coli by 70 percent and showing that a number of these regulatory elements have been conserved in bacteria from ancient ancestors while others have rapidly evolved. The new method, they say, is versatile enough to characterize genetic pathways across all kingdoms of life. Nature Methods
Subscribe to our free weekly newsletter.