First-Time Use of AI for Genetic Circuit Design Is Demonstrated in a Human Cell Line

There are hundreds of distinct cell types in the human body, and every one of them behaves the way it does because of instructions encoded in DNA. In theory, if scientists can write the right DNA sequences, they can program cells to do almost anything — produce therapeutic molecules, detect disease, or even assemble into replacement tissues. In practice, the challenge has always been figuring out which DNA designs produce which behaviors.

A new study from Rice University, published in Nature, shows that this long-standing problem in synthetic biology may finally be turning a corner. For the first time, researchers have successfully used artificial intelligence (AI) and machine learning to design and predict the behavior of genetic circuits in a human cell line, something that has never been achieved at this scale before.

At the center of this breakthrough is a new experimental platform called CLASSIC, short for Combining Long- and Short-range Sequencing to Investigate Genetic Complexity. The technique allows scientists to build, test, and analyze hundreds of thousands to millions of genetic circuit designs simultaneously, creating the kind of massive datasets that AI models need to work effectively.

Why Genetic Circuit Design Has Been So Hard

Genetic circuits are engineered combinations of DNA elements — such as promoters, transcription factors, and genes — that work together to control how a cell behaves. While scientists have known how to build simple circuits for decades, scaling up has been extremely difficult.

For any given biological function, there are enormous numbers of possible DNA configurations. Finding a design that works well has often felt like searching for a needle in a haystack. Traditional approaches rely on building circuits one at a time, testing them individually, and making small adjustments based on trial and error. This process is slow, labor-intensive, and poorly suited for exploring complex design spaces.

According to the Rice team, the core issue wasn’t a lack of ideas or theory — it was a lack of data at scale. AI models can only make reliable predictions when they are trained on very large, high-quality datasets. Until now, generating such datasets for complete genetic circuits simply wasn’t possible.

What CLASSIC Does Differently

CLASSIC changes the game by enabling ultra-high-throughput genetic circuit construction and testing. Instead of assembling a handful of circuits, the researchers developed a way to build huge libraries of DNA designs all at once.

The process begins with large-scale molecular cloning, where DNA is cut and reassembled in many different combinations. These combinations form vast libraries of genetic circuits, each slightly different from the others.

To analyze these circuits, the researchers used two types of next-generation sequencing:

• Long-read sequencing, which can read thousands to tens of thousands of DNA bases in a single pass. This method captures the entire genetic circuit, but it is slower and more error-prone.
• Short-read sequencing, which reads only a few hundred bases at a time but does so with high accuracy and speed.

Each genetic circuit was assigned a unique DNA barcode. Long-read sequencing was used to determine the full sequence of every circuit, while short-read sequencing tracked how often each barcode appeared during experiments. By combining both methods, the team could precisely link genotype (DNA sequence) to phenotype (how the circuit behaves in cells).

This combination of sequencing approaches is what gives CLASSIC its name — and its power.

Testing Genetic Circuits in Human Cells

To demonstrate the platform, the researchers built a proof-of-concept library of genetic circuits containing reporter genes that produce a glowing protein. These circuits were introduced into human embryonic kidney (HEK) cells, a widely used human cell line.

Once inside the cells, the circuits produced varying levels of fluorescence. Some cells glowed brightly, indicating high gene expression, while others glowed dimly. The researchers then sorted the cells into groups based on how strong the signal was.

Short-read sequencing of the DNA barcodes from each group allowed the team to reconstruct a detailed map showing how every individual circuit performed. This resulted in one of the largest and most comprehensive datasets ever created for genetic circuit behavior in human cells.

How AI Fits Into the Picture

Even with massive libraries, it is still impossible to experimentally test every possible DNA design. This is where AI and machine learning become essential.

Using the data generated by CLASSIC, the researchers trained machine learning models to learn the underlying rules that govern how genetic circuits behave. Once trained, these models were able to predict the performance of untested circuit designs with high accuracy.

To validate the approach, the team compared AI predictions against manual measurements from randomly selected circuits. The results matched closely — a key moment that confirmed the platform was working as intended.

This marks the first successful demonstration of AI-driven genetic circuit design in a human cell line, moving synthetic biology from handcrafted experimentation toward predictive, data-driven engineering.

Key Insights From the Study

Beyond the technical achievement, the study revealed several important biological insights:

• There is no single “correct” genetic circuit for a given function. Many different designs can produce similar outcomes.
• Circuits built from medium-strength components often outperform those using very strong or very weak elements. This suggests the presence of biological “Goldilocks zones”, where balance matters more than extremes.
• Machine learning models trained on large datasets were more accurate than traditional physics-based models at predicting circuit behavior.

These findings help clarify long-standing questions about how genetic parts interact in complex biological systems.

Why This Matters for Medicine and Biotechnology

Because CLASSIC was demonstrated in human cells, it has direct implications for real-world applications. Engineered cells are increasingly being explored as living medicines, capable of sensing disease states and responding dynamically.

Potential applications include:

• Cell-based cancer therapies
• Smart immune cells that adjust behavior in real time
• Biosensors that detect disease markers
• Programmable cells for tissue engineering and regenerative medicine

By dramatically accelerating the design-build-test-learn cycle, CLASSIC could significantly reduce the time and cost required to develop these technologies.

A Broader Shift in Synthetic Biology

Experts in the field have compared this work to earlier milestones such as the genetic toggle switch and the repressilator, which proved that cells could be programmed but were built slowly and individually. CLASSIC represents a shift in scale, allowing scientists to explore vast combinatorial spaces that were previously inaccessible.

Equally important is the collaborative nature of the project. The work brought together expertise from synthetic biology, physics, and computer science, including contributions from research groups at Boston University and Rice University’s computer science department. This cross-disciplinary approach reflects the direction modern biology is heading.

AI and the Future of Genetic Programming

The researchers believe that AI-driven design will become central to synthetic biology as datasets continue to grow. With more data, models can become more sophisticated, enabling the design of increasingly complex genetic programs.

Rather than replacing experimentation, AI acts as a powerful guide — helping scientists decide which designs are worth building in the first place. This partnership between wet-lab biology and machine learning could ultimately make cells as programmable as software.

Research paper:
https://www.nature.com/articles/s41586-025-09933-9