Researchers Use Large Genome Models to Design Synthetic Viruses

Researchers Use Genome Language Models to Design Novel Synthetic Viruses

Prof Marc Güell from the synthetic biology lab at Pompeu Fabra University in Spain stated that for the first time in history, biology is being designed on a computer. Prof Patrick Cai, chair of synthetic genomics at the Manchester Institute of Biotechnology, also called the study an important milestone that extends far beyond phages.

Viability and Diversity in AI-Generated Phages

According to Arstechnica, testing of the AI-outputted designs revealed varying degrees of success and structural uniqueness. Out of 285 outputs tested from a group, the overall fraction of viable viruses was 5.6 percent, representing 16 viable viruses. Outputs most similar to the original ΦX174 virus—specifically those with 98 percent sequence similarity and up—showed a viability rate of 46 percent.

View this post on Instagram about researchers large genome models, Researchers Use Large Genome Models

While the most effective viruses tended to closely resemble the original, individual AI-generated viruses exhibited distinct alterations:

• One virus lost a viral protein entirely while compensating through changes elsewhere in the genome.
• Another virus added an entirely new gene.
• Many featured genes that were longer or shorter than the original.
• One design replaced a ΦX174 gene with a gene originating from a distantly related virus.

Furthermore, none of the viable viruses featured a single change in the specific stretch of DNA where the replication of the virus genome starts.

Challenging Traditional Mutation Thresholds

Past studies have indicated that making a single base change altering just one amino acid in a viral protein carries an average 20 percent chance of inactivating the virus entirely. Based on mathematical models derived from that 20 percent inactivation probability, viruses with fewer than 25 changes theoretically possess only a 2.3 percent chance of viability, while probabilities above 25 changes drop to essentially zero according to standard calculations.

However, the AI-suggested designs defied these expectations. Out of roughly 300 viruses produced by the AI, five fell into the range of fewer than 25 changes, and three of those were viable. More notably, nearly a quarter of the AI-suggested viruses featuring more than 25 changes proved to be viable, which included two designs containing over 50 amino acid alterations.

Broader Implications and Future Horizons

Experts highlight that these computational advances could aid in tackling major challenges. Prof Güell noted that the technology allows researchers to dream of possibilities such as developing phages to tackle disease, enzymes to treat genetic disorders, and antibodies for immunotherapy.

Regarding future steps, researchers involved with the work noted that while generating living cells would require another significant leap—noting that the genetic code of the phage spans about 5,400 base pairs compared to at least 500,000 base pairs for the smallest living cell genome and three billion for the human genome—attempting simple organisms would be a lot of work but not impossible, and teams are interested in working toward that goal.