BioDiscoveryAgent is an AI agent designed to automate the process of designing genetic perturbation experiments, leveraging large language models (LLMs) and various tools. The agent aims to identify a subset of genes that, when perturbed, result in specific phenotypes, such as cell growth. By integrating biological knowledge and experimental outcomes, BioDiscoveryAgent can efficiently navigate the hypothesis space and improve upon existing methods. The agent's performance is evaluated on six datasets, demonstrating a 21% improvement in predicting relevant genetic perturbations and a 46% improvement in identifying non-essential gene perturbations compared to Bayesian optimization baselines. BioDiscoveryAgent also excels in predicting gene combinations, outperforming random baselines by more than twice. The agent's decision-making is interpretable, providing references to the scientific literature and allowing for human feedback. Overall, BioDiscoveryAgent offers a new paradigm in computational experiment design, enhancing the efficiency and interpretability of scientific discovery.BioDiscoveryAgent is an AI agent designed to automate the process of designing genetic perturbation experiments, leveraging large language models (LLMs) and various tools. The agent aims to identify a subset of genes that, when perturbed, result in specific phenotypes, such as cell growth. By integrating biological knowledge and experimental outcomes, BioDiscoveryAgent can efficiently navigate the hypothesis space and improve upon existing methods. The agent's performance is evaluated on six datasets, demonstrating a 21% improvement in predicting relevant genetic perturbations and a 46% improvement in identifying non-essential gene perturbations compared to Bayesian optimization baselines. BioDiscoveryAgent also excels in predicting gene combinations, outperforming random baselines by more than twice. The agent's decision-making is interpretable, providing references to the scientific literature and allowing for human feedback. Overall, BioDiscoveryAgent offers a new paradigm in computational experiment design, enhancing the efficiency and interpretability of scientific discovery.