Measuring AI Capabilities to Accelerate Biology Research

GPT-5 optimized a molecular cloning protocol, improving experimental efficiency by 79x compared to the baseline. OpenAI partnered with biosecurity startup Red Queen Bio to build a framework for evaluating whether AI can propose ideas, analyze results, and iteratively improve within a wet lab environment.

During the experiment, GPT-5 autonomously reasoned about the protocol and proposed modifications. In particular, it introduced a new mechanism combining two enzymes, RecA (from E. coli) and gp32 (from phage T4), to maximize performance. In this approach, gp32 cleans up the DNA ends, while RecA guides each strand to pair correctly.

Key experimental results are as follows:

• RAPF (RecA-Assisted Pair-and-Finish HiFi Assembly): 2.6x efficiency improvement over baseline
• T7 (Transformation 7): 36x efficiency improvement over baseline
• Combining the two protocols ultimately achieved a 79x performance improvement.

This research was conducted in a controlled environment to manage biosecurity risks. It demonstrates that AI can collaborate with human scientists to accelerate research speed, reduce costs, and help turn scientific discovery into real-world impact.