Topic
Computational method that predicts a protein's 3D structure from its amino acid sequence, widely used to infer function of uncharacterized proteins.
ITQB NOVA researchers engineered PsGalOx, which they call the first bacterial galactose oxidase, into a variant 40 to 1,800 times more efficient. The team says much of the gain in directed evolution came from cells making more protein; drug substrates are still ahead.
Reality
INRAE-led researchers modelled two lily moth odour receptors with AI and used them to find the species' previously unknown sex pheromone. The route could speed pheromone work on little-studied pests, though confirming the hit still took live moths.
Reality
An international collaboration added more than 8,000 predicted viral protein pairs to the AlphaFold Database, spanning about 2,800 human-infecting viruses. The models stop at pairs and leave out the sugar coats, so trimeric targets like the coronavirus spike are largely absent.
Reality
Stowers Institute scientists fed AlphaFold2 new aphid genomes and got about 2,400 structures of plant-hijacking proteins built on one reworked fold. Two X-ray structures anchor those predictions. What the proteins do inside a plant is still unknown.
Reality
A PNAS study compared predicted tail fiber structures for phages tested on Pseudomonas aeruginosa and found the broader-spectrum ones differ from each other where they touch the bacterial pilus. Dozens of database relatives fit the same pattern.
Reality
Apple researchers report that one end-to-end training run on more than 2 million sequence-and-structure pairs competes with the usual tokenizer-then-generator pipelines.
Reality
A team in Japan, Finland, Italy and the United States mapped ordered hydration shells above a peptide assembly at subnanometer resolution, and argues that water belongs in the definition of the protein.
Reality
NVIDIA backs its structure-prediction runtime with a production run across 4,777 proteomes rather than a benchmark chart, and the distance between that number and your cluster is mostly work BioIR never touches.
Reality
A group at the Institute for Molecular Science added a bias energy that pushes each new AlphaFold3 prediction away from the last, and the sampled set reached open, closed and intermediate states of a subunit that AF3 alone gets wrong.
Reality
NVIDIA's BioNeMo Agent Toolkit answers a real question for agents that know a task needs folding but not which model to call. The price of that answer is a database download and a Claude Science sandbox you have to open.
Reality
A Science paper from Stanford reports an AI-designed phage genome that produced a functioning virus. Predicting which phage infects which bacterium is still posed as an open question.
Reality