新興感染症の背後にある分子スイッチの解明Finding the molecular switches behind new infectious diseases

Researchers at Google DeepMind have described a method that uses artificial intelligence to pinpoint the molecular switches that regulate how emerging infectious disease pathogens behave. The work matters because the proteins that let bacteria, viruses, and parasites adapt to a human host, evade the immune system, or resist drugs often hinge on subtle structural changes that are difficult to observe experimentally. Identifying these switches could give researchers a faster route to understanding pathogen biology and to nominating new therapeutic targets.

At the center of the approach is the idea of a molecular switch: a region of a protein that can toggle between distinct shapes, or conformations, and in doing so turn a biological function on or off. Many disease-relevant proteins are not static. They shift between active and inactive states in response to binding partners, chemical signals, or changes in their environment. These dynamic sites, including allosteric pockets that sit away from a protein's main active region, are frequently the levers that control virulence. Capturing them has traditionally required painstaking laboratory techniques such as X-ray crystallography or cryo-electron microscopy, which can take months or years per target.

The DeepMind team's method appears to build on the company's structural biology tools, most notably AlphaFold, which predicts three-dimensional protein structures from amino acid sequences. The latest generation, AlphaFold 3, extended those predictions to complexes involving proteins together with small molecules, nucleic acids, and other partners, making it possible to model interactions rather than isolated chains. By analyzing predicted structures and the ways they may rearrange, the researchers say they can flag candidate switch regions that plausibly govern a pathogen's behavior, then prioritize them for experimental follow-up.

This is significant for emerging infectious diseases specifically because time is a constraint. When a new or rapidly spreading pathogen appears, scientists often lack detailed structural knowledge of its key proteins and have few validated drug targets to work from. Computational screening can help narrow an otherwise vast search space, suggesting which proteins and which sites are worth the cost of laboratory investigation. The stated goal is not to replace experiments but to guide them, reducing the number of dead ends and helping teams focus limited resources.

It is worth setting the work in a broader context. AlphaFold, developed by DeepMind, has become a widely used reference across the life sciences since its structure database was opened to researchers, and the 2024 Nobel Prize in Chemistry recognized computational protein structure prediction and design. DeepMind's sister company, Isomorphic Labs, was spun out to apply similar techniques to drug discovery in partnership with pharmaceutical firms. Other tools in the field, including RoseTTAFold from academic groups and a growing set of models for protein design and molecular docking, reflect a wider shift toward AI-assisted structural biology. Understanding protein motion and allostery, however, remains a harder problem than predicting a single folded shape, and this new method is positioned within that active area of research.

Several caveats apply. Predictions of structures and conformational switches are hypotheses that still require validation in the laboratory and, ultimately, in disease models. A candidate switch that looks compelling computationally may prove difficult to target with a drug, or may not behave as predicted inside a living cell. Turning a validated target into an approved therapy is a long process involving medicinal chemistry, safety testing, and clinical trials, so any acceleration at the discovery stage is only one part of a much larger pipeline. Details such as which pathogens were studied, how the switches were confirmed, and how the results compare with established methods will be important for assessing the method's reliability.

Even with those qualifications, the direction is notable. If AI systems can reliably highlight the regulatory sites that make pathogens dangerous, they could complement traditional structural biology and shorten the early phase of drug discovery for infectious diseases that currently receive limited attention. The practical impact will likely depend on independent replication, peer review, and how well the computational predictions hold up against experimental data over time.