model
AI system from Google DeepMind that predicts 3D protein structures from amino acid sequences, widely used across biology and drug research.
No evidence-backed relationships are recorded.
Tokyo Metropolitan Institute of Medical Science researchers report that RAB1A and RAB1B, two of about 60 human RAB proteins, tie OPTN to ATG9A vesicles. The link helps set where the membrane around a damaged mitochondrion begins to form, and labs can now test that step directly.
Reality
AlphaGenome Atlas answers every possible one-letter question in the human genome in advance, roughly a petabyte of predictions. Noncommercial access opened Tuesday, commercial use waits on Google Cloud, and the launch write-up carries no accuracy number.
Perspective Coverage
3 publishersReality
A new report combines a survey of 637 scientists with 15 million Gemini conversations. Its authors trace the gap between mathematics and biology to what it costs to check whether an answer is right.
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
GPN-Star folds whole-genome alignments and a species tree into its architecture, which is how it trains in hours where Evo 2 needed 2,000 processors and months. The cost of building those alignments sits outside that comparison.
Reality
Academics get all 9 billion single-base predictions through a browser at no cost, while commercial users wait on a Google Cloud licence whose terms DeepMind has not published, including for its own sister company.
Reality
Mutations in different autism genes break the same protein contact in UCSF's new Science atlas, and that observation is what ARIA is now paying $46 million to turn into a molecule. The organoid evidence sets the ceiling on what it can mean so far.
Reality
HydroGym ships more than 60 validated environments reaching Re = 4 x 10^5, plus a proof of concept where agents trained on cheap surrogates cut local skin friction on a 3D wing section.
Reality
The startup opened a robotic "human data center" near San Francisco on the thesis that AI drug discovery is limited by causal human experiments, not model capacity.
Reality
The startup opened what it calls the world's largest human data center last week. Its diagnosis of the field is testable; most of the evidence for its own machines is self-reported.
Reality
AIntibody tested 511 antibodies from 29 organizations on three tasks. Ranking clones inside HCDR3 clusters was worse than random for every model but one.
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