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

July 22, 2021

Protein folding race heats up

Last week we reported that the “RoseTTAFold” deep-learning algorithm from the University of Washington’s Institute for Protein Design in Seattle using open-source AI was finally giving Google DeepMind’s own AI-based protein folding algorithm “AlphaFold” some real competition. Now DeepMind is firing back. Today they made the stunning announcement in collaboration with the European Molecular Biology Laboratory that AlphaFold has predicted the structures of 98.5 percent of all the proteins in the human proteome—around 20,000 in total, most of which have no experimental structures reported to date. They created a public database to make the data freely available, and have already announced plans to predict in the coming months the structures of virtually every known protein in nature—more than 100 million in all. Does this remind anyone else of the Bezos–Branson race? Nature

A 3D model of a fruit fly protein.

A protein (UniProt Q9VZS7) from the fruit fly. DeepMind

Human gene increases rice and potato crop yields

Researchers at Peking University in Beijing have inserted a human gene that codes for the protein RNA demethylase FTO into rice and potato plants, creating transgenic crops that showed stimulated root proliferation, increased photosynthesis and drought tolerance, and up to 50 percent increases in crop yields for plants grown in the field. The transgenic expression of human FTO, originally identified as a fat mass- and obesity-associated protein, appears to increase the biomass of rice and potato crops by up to 50 percent. The researchers say this could be a promising strategy to increase agricultural productivity to address future needs where population growth, climate change, and agricultural pests may increasingly threaten food supplies. Hundreds of millions of people around the world are already undernourished, and last year was a wake-up call for addressing the world’s hunger needs, as supply chain and agricultural interruptions due to the pandemic caused tens of millions more to go without enough food to eat. Nature Biotechnology

Neural network improves prediction of drug metabolism

One of the ways precision medicine can improve cancer therapy is through pharmacogenomics, in which a patient’s genetic profile is used to guide drug choices and doses. Slight variations in the genes coding for certain enzymes associated with drug metabolism can translate into huge differences in how well a prescribed medicine works because some people clear the drugs faster than others. Scientists at Leiden University in the Netherlands have trained a neural network to make better sense of that variation by combining sequence data of a drug-metabolizing cytochrome P450 enzyme called CYP2D6 with data from 561 people showing how well they were able to metabolize the breast cancer drug tamoxifen. In a proof-of-concept study, they showed their approach accounted for 79 percent of variability in CYP2D6 activity between individuals compared to 54 percent with the conventional approach. Science Translational Medicine

What newborn genomic screening should look like

Genetic screening programs for newborns could identify hundreds of otherwise hard-to-diagnose diseases and save millions of lives. As genomic screening becomes more routinely available, there is no question these programs have potential breakthrough, population-wide benefits. The question is, how should newborn DNA sequencing data be used in a clinical setting? A recent review of 36 studies by Murdoch Children’s Research Institute and the University of Melbourne, Australia, suggests that genomic newborn screening programs should be developed and implemented in a nuanced fashion, incorporating questions of clinical feasibility as well as ethical, legal, and social considerations in order to address existing gaps in equity, access, education, and informed consent. JAMA

A high-density, low-cost flexible microprocessor

There’s nothing new about flexible electronics—but this one sounds really promising. Researchers at the British semiconductor company Arm have designed a 32-bit flexible microprocessor they call PlasticARM that uses metal-oxide, thin-film transistors less than one micron thick embedded onto a flexible substrate. They claim it’s a bona fide breakthrough because of the record density they achieved in logical elements—at least 12 times higher than previous state-of-the-art flexible integrated circuits. The materials involved should make these devices extremely cheap to produce, and they could be fabricated on a mass-scale. Their technology “will pioneer the embedding of billions of low-cost, ultrathin microprocessors into everyday objects,” the researchers predict. The market for wearables is probably just getting started. Nature

A new resource for studying sepsis and aging

Neutrophils are an abundant if short-lived immune cell, accounting for more than half of all the white blood cells in circulation at any given time. The bone marrow continuously releases them into the bloodstream, and they typically live half a day or so, playing a key role as a first line of defense against infection. But neutrophil dysfunction is implicated in sepsis as well as chronic inflammation as we age—the phenomenon known as inflamm-aging—making them a hot topic for research. Now a team of researchers at the University of Southern California has developed a large, multi-omics resource based on transcriptomic, metabolomic, and lipidomic data taken from the neutrophils of male and female mice, both young and old, which could help identify new targets for therapeutics. Nature Aging