In a controlled laboratory experiment, scientists fed a large‑scale genome‑modeling artificial intelligence system the DNA sequence of the bacteriophage ΦX174 and asked it to propose new viral designs. The algorithm produced about 300 candidate genomes, which the team then synthesized and tested for infectivity.
Out of the 300 attempts, only 16 viruses—just 5.6 percent—were able to replicate, confirming the long‑standing observation that ΦX174 is highly sensitive to genetic alteration. When the researchers narrowed the pool to those designs that retained at least 98 percent sequence similarity to the natural virus, the viability rate jumped to 46 percent, indicating that close resemblance to the original genome remains a key factor in functional virus creation.
All viable candidates retained the core set of genes found in the wild‑type ΦX174. None of them showed changes in the genomic region where the virus’s genome duplicates, a stretch previously identified as essential for replication. This consistency aligns with earlier work showing that a single amino‑acid substitution in many ΦX174 proteins carries roughly a 20 percent chance of rendering the virus non‑functional.
Despite the overall trend toward similarity, the individual AI‑designed viruses displayed a surprising range of differences. One variant lost an entire viral protein, compensating for the loss with multiple mutations elsewhere. Another added a completely new gene, while several others featured genes that were either elongated or truncated relative to the reference. In a striking case, a virus replaced one of its original genes with a segment taken from a distantly related phage, illustrating the model’s capacity to recombine genetic material across distant lineages.
The researchers performed a statistical analysis based on the 20 percent inactivation probability per amino‑acid change. Their calculations suggested that a virus with fewer than 25 amino‑acid alterations would have only a 2.3 percent chance of remaining viable. Of the AI‑generated set, five designs fell below that threshold, and three of those proved infectious. Conversely, viruses bearing more than 25 changes were expected to be dead ends, yet nearly a quarter of those heavily mutated designs still formed plaques, including two with over 50 amino‑acid substitutions.
These findings underscore both the robustness and the limits of ΦX174’s genome. While the phage tolerates only modest changes in most contexts, the AI’s ability to explore a vast mutational landscape uncovered rare, functional outliers that would be unlikely to arise by chance. The work highlights the need for continued vigilance in synthetic biology, as computational tools become increasingly adept at navigating the fine line between harmless experimentation and the creation of novel, potentially hazardous viral forms.
Este artigo foi escrito com a assistência de IA.
News Factory APP - notícias agênticas para impulsionar seu SEO e AEO.