> ## Content Index > Fetch the complete content index at: https://nexi.fund/llms.txt > Use this file to discover other available public pages before exploring further. # When AI Learns to Write the Code of Life - URL: https://nexi.fund/ai-genetic-circuits-cell-therapy-2026/ - Published: 2026-07-15T18:00:37.000Z - Updated: 2026-07-15T18:00:37.000Z - Description: First-time AI design of genetic circuits in human cells marks a new phase in synthetic biology and opens a path to truly programmable cell therapies. - Author: Nexi.fund Labs - Tags: AI & Infrastructure, #mode-3, #hook-paradox, #track-F Twenty-five years ago, two papers in Nature demonstrated something that seemed like science fiction: a genetic toggle switch that could flip between stable states inside a living cell, and a three-gene oscillator called the repressilator that made bacteria blink on a predictable rhythm. Those circuits were hand-built. Each one took months to tune. The DNA sequences that worked were discovered by trial and error, not by design. That era is ending. In January 2026, a team at Rice University published a technique called CLASSIC (Combining Long- and Short-range Sequencing to Investigate Genetic Complexity) in Nature. It is the first demonstration that AI can design genetic circuits in human cells, and it changes what is possible in synthetic biology at a foundational level. 🎯 For the first time, artificial intelligence can design genetic circuits in human cells. The Rice University CLASSIC platform generates and tests millions of DNA designs simultaneously, creating datasets large enough to train predictive AI models. The convergence of generative AI and synthetic biology has reached an inflection point: from the first genetic toggle switch in 2000 to CLASSIC in 2026, the field has moved from building one circuit at a time to exploring design spaces at industrial scale. The synthetic gene circuits market is projected to grow from $2.38 billion in 2026 to $4.64 billion by 2030 at an 18% CAGR, driven by AI-integrated design tools and demand for programmable cell therapies in oncology, immunology, and regenerative medicine. ## The CLASSIC Breakthrough From needle-in-a-haystack to industrial-scale genetic exploration The fundamental problem in genetic circuit design is combinatorial. A circuit of ten genetic parts (promoters, ribosome binding sites, coding sequences, terminators) has more possible configurations than a research team can build and test in a lifetime. Even a modest circuit with five regulatory elements and two output genes has tens of thousands of potential architectures. The standard approach has been to hypothesize a design, build it, test it, and iterate based on the results. A single design-build-test cycle can take weeks. Slow. Manual. Brittle. And fundamentally limited: the number of circuits you can test determines how much you can learn. If you only test a few hundred designs, you cannot build a predictive model that generalizes across architectures. The platform replaces that workflow with a parallelized pipeline. The team built libraries containing hundreds of thousands of complete genetic circuits, pooled them, and inserted the entire collection into human cells in a single experiment. They then used a combination of long-read and short-read DNA sequencing to create a master map linking each circuit's DNA sequence to its measured behavior in the cell. The result is a dataset of unprecedented scale: millions of data points connecting sequence to function. That dataset is what makes AI possible. Without it, the models have nothing to learn from. "Our work is the first demonstration you can use AI for designing these circuits," Caleb Bashor, deputy director of the Rice Synthetic Biology Institute and senior author on the study, said. The team trained machine learning models on the data and found that the models could accurately predict the behavior of genetic circuits they had never tested. ## From Circuits to Generative Design The second inflection point in 25 years The technique is a proof of concept, but it is not an isolated result. The same month the Rice paper appeared, James Collins, one of the original inventors of the toggle switch, published a perspective in Cell Systems laying out a roadmap for generative AI in synthetic biology. The paper, co-authored with researchers at MIT and the Wyss Institute, argues that generative models can now design not just individual proteins but entire genetic circuits and, eventually, synthetic genomes. Collins called it a "transformative leap" that "redefines the scale" of what is possible in synthetic biology. The distinction matters: this is not an incremental improvement in an existing method. It is a new category of tool, one that replaces manual design with data-driven prediction. In March 2026, Nature published additional work on synthetic circuits for cell ratio control: recombinase-based devices that enable precise control over the ratios of cell types in an offspring population. In June, a review in Trends Open mapped the synthetic gene circuit field for cell-based precision therapy. The field is generating new results faster than any single lab can track. ## How Genetic Circuits Learn The data pipeline that makes AI-driven design possible The CLASSIC method works in four stages. First, the team designs a library of genetic constructs, each containing different combinations of promoters, coding sequences, and regulatory elements arranged in a circuit architecture. These are synthesized in parallel and pooled together. Second, the pooled library is inserted into human cells, where each construct lands in a different cell and begins to function. Third, the researchers extract DNA from the entire cell population and sequence it using two complementary methods. Long-read sequencing reads entire circuit constructs end-to-end, identifying which DNA sequence is present. Short-read sequencing quantifies how much RNA each construct produced, which is the functional output. By aligning the two datasets, the team builds a map linking every DNA sequence to its measured activity level. Fourth, that map becomes training data for machine learning models. The models learn the rules that relate sequence to function, then predict the behavior of new, untested circuit designs. The models outperformed traditional physics-based simulations, which require detailed mechanistic assumptions about each genetic part. This is important because the design space is enormous. With CLASSIC, the Rice team generated measurements for hundreds of thousands of designs in a single experiment. That is more data than the entire field of synthetic biology had accumulated in the two decades before 2026\. The models trained on this data did not just memorize the tested circuits. They learned the underlying grammar of how sequence context affects behavior, enabling them to predict entirely new configurations. ## Beyond CAR-T: The Next Generation Logic gates, closed loops, and feedback control in living cells The SynNotch AND gate is just the beginning. More complex circuits now implement closed-loop control: a therapeutic cell monitors a disease biomarker in the patient's bloodstream and adjusts its drug output in real time, increasing or decreasing production as needed. This is not a single-dose therapy. It is a living feedback system. Closed-loop metabolic circuits represent a parallel therapeutic frontier. Researchers have engineered cells that sense elevated levels of metabolites, such as uric acid or phenylalanine, and respond by producing enzymes that break them down. These circuits are being developed for metabolic disorders like gout and phenylketonuria, where current treatments require lifelong dietary restriction or frequent drug administration. Strand Therapeutics is engineering programmable mRNA therapies with gene circuits that switch on only in targeted cells. Their technology uses synthetic biology to control mRNA expression spatially and temporally, minimizing off-target effects. Outpace Bio applies AI-driven protein design to cell therapy, building T cells with enhanced persistence and tumor microenvironment resistance. Its OutLast and OutSmart platforms represent a convergence of protein engineering and synthetic circuit design. Another area of active development is cell ratio control, engineering a population of cells to maintain a specific ratio of different subtypes. This matters for therapies that require multiple cell types working together, such as a combination of killer T cells and helper T cells in a single therapeutic product. The Nature paper from March 2026 on synthetic circuits for cell ratio control demonstrates that bacteria can be programmed to maintain user-defined ratios across generations, pointing toward similar capabilities in therapeutic cell products. ## Programmable Cell Therapies Living drugs that make conditional decisions The most immediate application is in cell therapy. Today's CAR-T cells are programmed with a single receptor: if the target antigen is present, the cell kills. But cancers are heterogeneous. A single-antigen gate misses cells that downregulate the marker, and it cannot distinguish between tumor tissue and healthy tissue expressing the same antigen at low levels. Synthetic gene circuits solve this with logic-gated activation. The SynNotch system, developed at UCSF, enables an AND gate architecture: a T cell kills only when it detects two antigens simultaneously. Neither alone is sufficient. This is the difference between a simple tripwire and a conditional decision. Senti Biosciences, founded by Tim Lu, is building a platform around these principles, designing genetic circuits that make cell therapies smarter, with external controllability and context-dependent activation. Asimov, another startup in the space, offers a computer-aided design platform for mammalian synthetic biology, integrating computational modeling, machine learning, and multi-omics measurement into a single workflow. This is already the third AI-biology convergence story in our tracker this quarter. As we wrote last week, [AI-powered organ-on-chip platforms](https://nexithon.com/blog/xellar-ai-organ-chip-drug-discovery-2026/?ref=nexi.fund) are reshaping drug discovery — the same pattern of AI tools moving into biological wetware. The pattern is becoming clear: the same AI tools that transformed protein design are now being applied to the broader challenge of programming cellular behavior. The first wave was proteins. This wave is circuits. ## The Market Signal $2.38 billion and growing at 18% annually The synthetic gene circuits market was valued at $2.02 billion in 2025 and reached an estimated $2.38 billion in 2026, growing at 17.9% CAGR. Projections put it at $4.64 billion by 2030\. The primary growth driver is the integration of AI into circuit design, a category that barely existed before 2026. North America accounts for the largest regional share. Asia-Pacific is the fastest-growing market, driven by biotech expansion in China, South Korea, and Singapore. Applications span healthcare, industrial biotechnology, R&D, and agriculture. The therapeutic segment (cell and gene therapy) is the most valuable and the fastest-growing. The numbers reflect a structural shift, not a cyclical one. Precision medicine requires programmable biology. Programmable biology requires genetic circuits. And designing genetic circuits at scale requires AI. That is the chain. The links are now testable. It did not exist a year ago. Each link in it represents a market opportunity that is only now becoming visible to institutional investors who track the infrastructure layer of the AI-biology convergence. ## The Investment Opportunity Platform companies, tool providers, and the enabling layer The most investable opportunities in AI-driven genetic circuits are not therapeutic candidates. They are platform companies that own the design tools. Asimov, with its computer-aided design platform for mammalian synthetic biology, occupies a position analogous to what Cadence and Synopsys built for semiconductor design in the 1980s. If programming cells becomes a standard engineering practice, the companies that own the design environment will capture disproportionate value. Senti Biosciences and Strand Therapeutics represent the therapeutic application layer. Their value depends on clinical execution rather than platform adoption, which introduces binary risk: a failed Phase 2 trial can wipe out years of progress. But successful clinical validation of a logic-gated CAR-T product would create a strong moat, because the regulatory pathway for a programmable cell therapy is harder to replicate than the genetic circuit itself. At the tool layer, companies like Twist Bioscience and Ginkgo Bioworks supply DNA synthesis and strain engineering services that the circuit-design pipeline depends on. As CLASSIC-like approaches scale, demand for high-throughput DNA synthesis will grow. The companies positioned to benefit are those that can deliver large, complex libraries quickly and at low cost. The academic ecosystem is also a source of deal flow. The Rice team's CLASSIC platform was developed with NIH and Office of Naval Research funding. University technology transfer offices are already fielding inquiries from venture capital firms interested in exclusive licenses to the underlying methods. The first startup spun out of this specific work has not been announced, but the pattern of academic breakthrough, venture funding, and clinical pipeline is well established in synthetic biology. ## The Dual-Use Dimension AI-designed biology raises questions the field has not answered Every capability in synthetic biology is dual-use. The same design tools can program therapeutic cell functions or harmful biological functions. The underlying logic is identical. In October 2025, a team at MIT published a review of AI-designed protein threats, finding that existing DNA synthesis screening tools may not catch sequences ordered from AI-generated designs. The concern is not hypothetical: several commercial DNA synthesis companies have already faced orders for toxin-coding sequences designed by generative AI models. CLASSIC operates at a different layer. It designs the logic of genetic circuits rather than the sequence of individual proteins, but the same governance gap applies. A toggle switch that controls therapeutic gene expression and a toggle switch that controls toxin production use the same design principles. The pipeline that generates and tests millions of circuits can explore harmful designs just as efficiently as beneficial ones. The synthetic biology community is aware of this. The Collins perspective in Cell Systems explicitly calls for a governance framework for AI-designed biological systems. The US National Security Commission on Emerging Biotechnology has recommended that DNA synthesis screening be made mandatory. No federal mandate exists yet. The field is self-governing through the International Gene Synthesis Consortium, which screens orders on a voluntary basis, but its members represent a fraction of global synthesis capacity. For investors, this is not a reason to avoid the space. It is a factor to price in. Regulatory frameworks will eventually arrive, and companies that have already built strong biosecurity practices will face less disruption than those that ignore the issue. The FDA has not yet published guidance specifically for AI-designed genetic circuits in cell therapy products, but the agency's 2025 discussion paper on AI in drug development signals that a framework is under development. 📊 **Key signals to track** Clinical entry of first logic-gated CAR-T therapy, such as SynNotch or similar AND-gate architecture reaching Phase 1 or 2 Asimov or Senti Bio platform licensing deal with a top-10 pharmaceutical company CLASSIC-like technique applied to non-human cell types for industrial biotechnology FDA regulatory framework for AI-designed genetic circuit safety characterization ## Sources [ Ultra-high-throughput mapping of genetic design space The CLASSIC paper, the first demonstration of AI-designed genetic circuits in human cells. Published in Nature, January 2026. Nature ](https://www.nature.com/articles/s41586-025-09933-9?ref=nexi.fund) The primary research result: CLASSIC enables million-scale genetic circuit libraries with AI-driven analysis [ Designing Programmable Cancer Therapies with Synthetic Gene Circuits Overview of synthetic gene circuits in therapeutic applications, covering SynNotch logic gating, Senti Biosciences, and the clinical translation landscape. SynBioBeta, 2024. SynBioBeta ](https://www.synbiobeta.com/read/designing-programmable-cancer-therapies-with-synthetic-gene-circuits?ref=nexi.fund) Industry publication covering the therapeutic application layer of synthetic gene circuits [ Houston scientists develop breakthrough AI-driven process to design, decode genetic circuits Detailed coverage of the CLASSIC technique and its implications from InnovationMap, February 2026. InnovationMap ](https://houston.innovationmap.com/rice-university-classic-genetic-circuit-2675264254.html?ref=nexi.fund) Background reporting on the Rice team's four-year development of the CLASSIC platform