OpenAI and Ginkgo Bioworks: AI Revolutionizes Biological Research

Technological advancements continue to amaze, and the recent collaboration between OpenAI and Ginkgo Bioworks is a perfect example. These two entities have demonstrated that artificial intelligence, combined with a fully automated laboratory, can accelerate the design and improvement of biological experiments at an unprecedented pace.

The initial idea was simple: could OpenAI's GPT models, already capable of summarizing scientific studies and making predictions, also formulate scientific hypotheses, design experiments, analyze results, and adjust their approach? This question is particularly relevant in the field of biology, where answers are not always straightforward. Joy Jiao, head of life sciences research at OpenAI, emphasizes the complexity of generating and verifying a hypothesis.

An Innovative Collaboration

Last summer, OpenAI decided to test this idea with Ginkgo Bioworks, a company specializing in automated laboratories. OpenAI's GPT-5 model was used to draft experimental protocols, while Ginkgo Bioworks set up an automated system that its CEO, Jason Kelly, compares to the "Waymo" of biology.

Researchers focused their efforts on the superfolder green fluorescent protein (sfGFP), a lab-modified protein used as a reference for its ability to produce detectable green light. The GPT-5 model, from OpenAI's headquarters in San Francisco, sent its protocols to Ginkgo Bioworks' robots in Boston, allowing the AI to manage experiments without constant human oversight.

Experiment Methodology

To test the capabilities of the AI, researchers relied on the superfolder green fluorescent protein, or sfGFP. This protein, derived from a jellyfish and modified in the lab, is frequently used as a reference because it produces easily detectable green light, allowing for quick assessments of whether an experiment is successful.

The work focused on cell-free protein synthesis, also known as CFPS. This technique allows for protein production without using living cells, thus avoiding several lengthy steps, such as inserting DNA into a cell or multiplying it.

Impressive Results

In two months, over 36,000 tests were conducted, and the AI managed to reduce the cost of protein production by approximately 40%. Initially skeptical, researchers were surprised by the results: the experiment produced a measurable amount of proteins, exceeding expectations.

Michael Jewett, a bioengineer at Stanford University, believes that this combination of AI and autonomous laboratories could accelerate the development of new drugs. Reshma Shetty, co-founder of Ginkgo Bioworks, emphasizes that the potential of AI models is fully realized when they are paired with laboratories capable of testing their ideas in the real world.

An unexpected moment occurred when GPT-5 attempted to use a negative amount of water in its calculations, a situation that is practically impossible. Technicians adjusted the volumes to conduct the tests.

Towards a Promising Future

Today, the AI-optimized reaction composition is commercially available. On March 2, Ginkgo Bioworks launched Ginkgo Cloud Lab, a platform allowing researchers to submit their experiments to autonomous laboratories, starting at $39 per analysis.

This approach is revolutionizing biological research by significantly accelerating the pace of experiments. Where a traditional laboratory would take several days to complete a full cycle, AI and robots can accomplish this in about an hour, testing thousands of variants in just a few weeks.

This innovation also reduces costs, which is crucial for the pharmaceutical and biotechnology industries. With only 12% of molecules tested in Phase I becoming approved drugs, the potential savings are enormous. According to Deloitte, developing a drug costs approximately $2.23 billion for large pharmaceutical companies in 2024. Reducing research timelines or avoiding unnecessary experiments could lead to significant savings.