Artificial intelligence is making rapid strides in computational biology. Recently, researchers at Stanford and the Arc Institute created the world’s first entirely AI-generated genome. In what can be called a breakthrough, the new virus created by AI can infect and kill bacteria.
Scientists have already used AI to design individual proteins and even small multi-gene systems. However, creating an entire genome is way more complex. In simple words, a genome must have many interacting genes, and regulatory switches that allow an organism to grow, copy itself, and survive. For scientists, until now, getting all of these to work together was a huge challenge.
“Genome design requires orchestrating multiple interacting genes and regulatory elements while maintaining a balance that enables replication, host specificity, and evolutionary fitness. This increase in complexity introduces new constraints and failure modes that do not arise when only designing a single protein or a two-component system,” the team said in a post on arcinstitute.org.
How did they do it?For the test case, the researchers selected a tiny virus named bacteriophage ΦX174 (pronounced phi-X-174). This virus infects E. coli bacteria and has a tiny yet tricky genome – 5,386 letters of DNA and 11 genes, most of which overlap each other. This virus was also the first genome ever to be fully sequenced in 1977 and the first ever synthesised from scratch in 2003. Now, it is the first to be designed by AI.When it came to training the AI, the scientists used a genomic language model named Evo which was fine-tuned on thousands of genomes from the virus’s family so it could speak the dialect of ΦX174.
With the help of prompts, the AI-generated thousands of candidate genomes. The team did a series of quality checks and lab tests. They built a custom software to ensure that each design consisted of all key genes and essential proteins required to infect E.coli. Later, the team synthesised hundreds of these AI genomes in the lab and tested them by inserting them into bacteria to see if they could reproduce. Following this, 16 new functional viruses emerged and each of them carried dozens to hundreds of mutations never seen in nature. One standout design even borrowed a DNA-packaging protein from a distant relative, something human engineers had tried and failed to do before. Cryo-electron microscopy confirmed the novel protein fit and worked inside the virus shell.The scientists had trained EVO on 2M viruses and later asked them to design brand new ones. Out of 302 attempts, 16 proved to be functional in lab tests.
The AI viruses contained 392 mutations never seen in nature including combinations that scientists had previously tried and failed to create. When the bacteria developed resistance to natural viruses, AI-designed versions passed through defenses in days where traditional viruses failed. This is a significant achievement as it marks a new phase in biotechnology.
From reading DNA (sequencing), to writing it (synthesis) and now to design it, this is a new milestone. At a time when AI is redefining domains like education, productivity, and various avenues of creativity, a devlopment like this shows how far we have come. The biggest takeaway here is the boundless potential AI holds for accelerating scientific discoveries and breakthroughs.
