The study introduces URSA (Utilitarian RetroSynthesis Assessment), an evaluation framework designed to benchmark synthetic routes from both formal and chemical plausibility perspectives. This protocol addresses the difficulty of objective comparison in drug discovery by mimicking how expert chemists evaluate reactions.
- The framework assesses convergence to commercially available starting materials alongside chemical plausibility.
- It evaluates conventional end-to-end retrosynthesis solutions and large language models on novel, diverse target molecules with undisclosed synthetic routes.
- LLMs are found to support high-level strategic planning but underperform specialized retrosynthesis models in reliably solving synthesis planning tasks.