Mutating every DNA letter of a genome shows the limits of AI (www.nature.com)

A recent study on the phage ΦX174, a well-characterized virus with a circular single-stranded DNA genome, revealed significant limitations of current AI models in predicting the effects of genetic mutations. Researchers systematically mutated nearly every nucleotide and amino acid in the virus’s genome, leading to over 44,000 variants. The findings showed that around half of the single-nucleotide mutations and 60% of the amino acid changes were detrimental to the virus's survival, yet the underlying reasons for many of these harmful effects remain unclear. This research highlights the intricate complexities of viral genetics and emphasizes the necessity for improved experimental data to enhance AI tools in biological research. While cutting-edge AI systems have been effective in identifying potential harmful mutations, their struggle with such a well-studied organism underscores the challenges of accurately modeling biological systems. The study contributes to our understanding of molecular interaction dynamics and suggests that even optimized biological models like phi X have room for improved adaptability, further motivating AI advancements in this field.