Reviewer #2 (Public review): Benchmarking biochemical networks generated by large language models

The authors propose a pipeline using large language models (LLMs) to benchmark experimentally well validated computational models of signaling networks.The findings are important, with available methods that guide testing future models and find new molecular interactions.Furthermore, the work shows how general-purpose LLMs can generate up to 91% of reactions of a bacterial metabolic network.The support is convincing and offers a number of performance metrics.