AI's Limits in Enzyme Function Prediction: A Nature Paper's Hidden Errors

A Nature paper used a Transformer model to predict the function of 450 unknown enzymes, garnering significant attention. However, a subsequent paper revealed hundreds of errors in these predictions. This highlights the limitations of AI in biology and the flaws in current publishing incentives. Careful examination showed many predictions weren't novel, but were repetitions or outright incorrect. This underscores the importance of deep domain expertise in evaluating AI results and the need for incentives focused on quality over flashy AI solutions.