AI Genome Models Yield Viable Engineered Viruses in Lab Tests

Key Points

• AI model generated ~300 ΦX174 genome variants for lab synthesis.
• Only 16 designs (5.6%) produced viable, replicating viruses.
• Viability rose to 46% among designs with ≥98% sequence similarity to the wild type.
• All functional viruses kept the original gene set and unchanged duplication region.
• Some variants lost or added genes, and one swapped a gene with a distant phage.
• Statistical model predicted low success for >25 amino‑acid changes, yet several heavily mutated viruses remained viable.
• Results illustrate both the resilience of ΦX174 and the power of AI in exploring viral mutational space.