Mutating every DNA letter of a genome shows surprising effects — and the limits of AI

Nature NewsIndependentCenteryesterday

A study led by Ben Lehner and colleagues systematically mutated nearly every nucleotide in the well-studied bacteriophage ΦX174 genome, revealing significant gaps in understanding mutation effects despite its extensive prior research. Over 44,000 variants were tested by culturing them with Escherichia coli to assess their viability, finding that over half of single-nucleotide mutations and 60% of amino acid alterations were harmful. While some mutations unexpectedly enhanced the virus's fitness, advanced AI models failed to accurately predict most outcomes, highlighting the need for more robust experimental data to improve biological AI tools. This work builds on earlier AI-designed versions of ΦX174 and underscores ongoing challenges in predicting mutation impacts.

Bias read (Center): The article presents scientific findings without overt ideological framing. It discusses technical limitations of AI in predicting mutation effects and emphasizes the need for better data, without taking a partisan stance. The focus remains on empirical research and technological development rather