The week’s most astounding developments from the neobiological frontier.

July 13, 2023

Like Dall-E for protein design

Efforts to solve the 50-year-old “protein folding problem” have borne fruit in recent years, thanks to AI. It’s an important, if inscrutable challenge: Understanding how human proteins fold and being able to predict their exact shapes in different disease states—which is often the starting point for drug design. In the last three years, the deep learning algorithms of the University of Washington’s RoseTTAFold and Google DeepMind’s AlphaFold2 have been both hugely successful and highly competitive with each other at predicting the structures of proteins. Now researchers at UW have taken the next step, developing the first generative AI tool for protein design, which they are calling RoseTTAFold Diffusion (RFdiffusion). Similar to image creation programs like Dall-E, the new tool could open up whole new worlds for protein design, promising “to approach and—in some cases—surpass what natural evolution has achieved,” the researchers write. Nature

Predicting human error based on brain waves

Precise sequential motor actions are part of everyday life, and our ability to execute them allows us to perform complicated, controlled maneuvers, from acing a serve in tennis to landing a jumbo jet on an airstrip. In the case of landing an airplane, the impact of poorly executing sequential actions is huge and potentially disastrous. Asking whether they could predict such errors before they occur, researchers at the National Institutes of Health in Bethesda, Maryland, used magnetoencephalography to observe the brain waves of people performing a challenging sequential task. In the moments before a test subject made a mistake, they showed anomalous low-frequency oscillatory brain activity—similar to what researchers expect to see during sleep or states of low consciousness. Those signals allowed the researchers to predict with 70 percent accuracy when their test subjects were about to err. Current Biology

AI for fighting future infections

A nice review this week by researchers at MIT and the University of Pennsylvania on using AI to fight infectious diseases appears in a special collection of articles in Science focused on the future of AI in health care and biomedical research. In the review, the authors see a bright future for AI as a tool for understanding and fighting infectious diseases, suggesting it could help control emerging outbreaks and curtail future pandemics by enhancing the design of next-generation drugs, vaccines, and diagnostic tools. Science

People who participate in genetic studies are genetically inclined to do so

A significant challenge in doing genetics research, taking political polls, or doing any other types of representative population sampling is a phenomenon known as “participation bias.” When people holding a specific political position, for instance, are more motivated to participate in a poll, their overrepresentation can skew the results. Likewise in genetics studies, if the participants disproportionately possess specific traits or genes, they could appear more common than they actually are. Now researchers at the University of Oxford have discovered an unusual feature to this type of bias: People who participate in genetic studies are genetically more likely to do so. Using data from the UK Biobank, they identified a genetic component to participation, which they report is responsible for leaving detectable “footprints” in their genetics data. This work should help researchers identify and eliminate such participation bias in future research. Nature Genetics

The beginning of the Anthropocene—or the end of the discussion?

Many hold that we have entered the so-called Anthropocene epoch, a new era in Earth’s geological history in which human activity has become the dominant influence on the planet. But experts still debate whether the Anthropocene has begun or not. It’s up to a body of experts known as the International Commission on Stratigraphy to decide, and they have been collecting drill core samples around the world—from tropical coral reefs to arctic ice sheets—to look for heavy metals, coal ash, radioactive plutonium, and other environmental evidence of human activity. This week, the commission announced its primary reference spot where the question will be decided: Crawford Lake, outside Toronto. Sediments found at the bottom of this lake provide “an exquisite record of recent environmental change over the last millennia,” a member of the commission said this week. Press announcement

Annual sediment samples from the Crawford Lake site. University of Southampton