What happened: Stanford–Arc Institute genome “language models” generate whole phage genomes that are synthesised and tested

A Stanford–Arc Institute study used artificial intelligence to shift biology from analysing genetic information to proposing complete genome designs that researchers can build in a laboratory. The study trained genome “language models” on large genomic datasets, especially bacteriophage genomes related to the well-studied virus ΦX174.

The genome language models generated new, previously unseen whole-genome designs within a known biological framework. Researchers then physically synthesised selected designs and tested them in bacteria to check whether functioning phages formed.

Experimental results showed mixed performance. Only a minority of the designed genomes produced functioning phages. However, several designed phages were able to overcome bacterial resistance that had defeated the original phage used as the reference framework.

Background: AI in biology has often helped analysis rather than direct DNA design

Before the reported approach, many practical AI uses in biology focused on analysing data, predicting properties, or interpreting patterns in existing sequences. The Stanford–Arc Institute work emphasises a different capability: design generation, where the model outputs candidate whole-genome DNA designs that can be taken to the wet lab.

Related current affairs