GPT-5 Lowers the Cost of Cell-Free Protein Synthesis

OpenAI and Ginkgo Bioworks have demonstrated that GPT-5 can reduce the cost of cell-free protein synthesis (CFPS) by 40% and reagent costs by 57% through closed-loop autonomous experimentation. By connecting the model to a cloud laboratory, the system established a new state of the art in low-cost protein production within three rounds of experimentation.

Autonomous Optimization of Cell-Free Protein Synthesis

Cell-free protein synthesis (CFPS) allows for the production of proteins without the need to grow living cells by running protein-making machinery in a controlled mixture. This process is critical for the production of medicines, diagnostics, research assays, and industrial enzymes, but it is traditionally difficult to optimize due to the complex interactions between DNA templates, cell lysates, and various biochemical components.

To address these bottlenecks, OpenAI paired GPT-5 with Ginkgo Bioworks’s cloud laboratory—an automated wet lab run remotely via software. This created a closed-loop system where GPT-5 designed experiment batches, the robotic lab executed them, and the resulting data was fed back to the model to inform the next round of hypotheses.

Technical Execution and Scale

The system employed a strict programmatic validation layer to ensure that AI-designed experiments were physically executable on the automation platform, preventing the proposal of "paper experiments" that cannot be carried out by robots.

Over six rounds of experimentation spanning two months, the system achieved the following:

• Total Reactions: More than 36,000 unique CFPS reaction compositions tested.
• Total Plates: 580 automated 384-well plates.
• Timeline to SOTA: GPT-5 established a new state of the art in low-cost CFPS in three rounds of experimentation after gaining access to a computer, web browser, and relevant research papers.
• Cost Reduction: A 40% reduction in protein production cost and a 57% improvement in reagent costs compared to the best prior baseline.

Key Findings and Technical Insights

The autonomous system identified low-cost reaction compositions that had not been previously tested by humans, leveraging high-throughput capabilities to explore a vast space of possible mixtures.

Robustness to Automation Constraints

GPT-5 identified reagent combinations that were specifically robust to the conditions of high-throughput automation, which often differ from manual bench-top experiments. Specifically, the model found compositions that performed well in low-oxygen conditions common in microtiter plates, where oxygenation and mixing are typically lower than in test tubes.

High-Leverage Optimization Parameters

The research revealed that small changes in the following areas had an outsized impact on cost relative to their expense:

• Buffering
• Energy regeneration components
• Polyamines

Additionally, the team found that because costs in CFPS are dominated by lysate and DNA, increasing protein yield per unit of expensive input is the most effective strategy for reducing overall production costs.

Limitations and Future Work

The results were demonstrated using one protein (sfGFP) and one CFPS system. Further research is required to determine if these gains generalize to other proteins and systems. Additionally, while the model can design and interpret experiments, human oversight remains necessary for protocol improvements and reagent handling.

OpenAI intends to apply this lab-in-the-loop optimization to other biological workflows. The lab also noted that these capabilities have biosecurity implications, which are being assessed and mitigated through OpenAI’s Preparedness Framework.