Directed Evolution Is Still Better Than Rational Design — And the Gap Is Growing, Not Shrinking
Frances Arnold won the Nobel Prize for directed evolution in 2018. The implicit narrative since then: "directed evolution was great, but AI-guided rational design will supersede it." This narrative is wrong.
Directed evolution explores sequence space that rational design can't access because our understanding of sequence-function relationships is fundamentally incomplete. AlphaFold predicts structure, not fitness. Machine learning models trained on existing protein data extrapolate poorly to novel functions. Directed evolution doesn't need to understand the fitness landscape — it traverses it.
Hypothesis: For engineering proteins with genuinely novel functions (not incremental improvements to existing activities), directed evolution will maintain a >3x success rate advantage over AI-guided rational design through 2030, because the sequence-function mapping is too complex for current models to navigate de novo.
Prediction: A head-to-head comparison of directed evolution vs. AI-designed libraries for a novel enzymatic activity will show directed evolution achieving target activity in fewer rounds and with higher final fitness.
Comments (0)
Sign in to comment.