AI-Designed Proteins Could Reveal Previously Hidden Activity Inside Living Cells

Researchers developed NovoTags, a new family of AI-designed proteins that bind to bright fluorescent dyes and help scientists identify multiple proteins inside living cells. The technology could improve multicolor microscopy, reveal protein interactions, and connect fluorescence imaging with cryo-electron tomography.

Editorial Note

This article provides independent biotechnology and science reporting for educational purposes. It does not provide medical advice or suggest that NovoTags are approved diagnostic tests, treatments, or clinical products.

New To Education is not affiliated with, sponsored by, endorsed by, or acting on behalf of the University of Washington’s Institute for Protein Design, Howard Hughes Medical Institute, Janelia Research Campus, the European Molecular Biology Laboratory, Science, or any researcher or organization discussed in this article.

The research describes a laboratory imaging technology intended primarily for biological research. Although NovoTags could eventually help scientists study disease mechanisms and identify new therapeutic targets, the current work does not show that the technology can diagnose disease, treat patients, or directly improve clinical outcomes.

Scientists Have Created New Tools for Seeing Inside Living Cells

Researchers have developed a new class of fluorescent protein tags designed with artificial intelligence to help scientists observe biological activity inside living cells.

The tools, known as NovoTags, are small synthetic proteins created from scratch rather than modified from proteins already found in nature. Each NovoTag is designed to bind tightly and selectively to a particular fluorescent dye.

Once attached to a protein or cellular structure, the tag allows researchers to illuminate and track that target through advanced microscopy.

The collaboration brought together researchers from the Baker Lab at the University of Washington’s Institute for Protein Design, the Lavis Lab at the Howard Hughes Medical Institute’s Janelia Research Campus, and the Mahamid Group at the European Molecular Biology Laboratory in Heidelberg.

The peer-reviewed research was published in Science on July 16, 2026. EMBL publicly highlighted the findings on July 17.

The development could expand the number of proteins scientists can observe simultaneously and may eventually help researchers study interactions that have remained difficult to capture with existing imaging systems.

Cells Are Crowded and Constantly Changing

A living cell contains an enormous number of molecules performing different tasks at the same time.

Proteins transport materials, transmit signals, repair damage, produce energy, copy genetic information, control cell division, and respond to changes in the surrounding environment.

These activities do not happen in isolation. Proteins move between cellular compartments, join temporary complexes, separate again, and respond to other molecules within fractions of a second.

Scientists can often study one protein in detail, but understanding how dozens of proteins interact inside the same living cell is much harder.

Traditional laboratory methods may require cells to be broken apart or chemically fixed. Those techniques can provide valuable information, but they may also remove the movement and timing that make living biological systems so complicated.

Fluorescence microscopy helps solve part of that problem by allowing scientists to attach glowing markers to selected molecules and watch them inside cells.

The limitation is that researchers can only distinguish a restricted number of markers at one time.

NovoTags were developed to expand that visual vocabulary.

How Fluorescent Protein Tags Work

A fluorescent tag acts like a molecular label.

Researchers genetically attach the tag to a protein they want to study. A matching fluorescent dye is then introduced into the cell.

When the dye binds to the tag and the sample is exposed to the correct wavelength of light, it produces a visible signal. Scientists can then locate the tagged protein and track where it moves.

Existing systems such as HaloTag and SNAP-tag have become widely used in biological research. They allow scientists to combine engineered protein tags with bright synthetic dyes.

Those technologies have supported major advances in cellular imaging, but they have practical limits.

Several existing tags rely on similar chemical attachment systems. That makes it difficult to label many proteins independently without the dyes or connectors interfering with one another.

Even when different colors are available, their emission patterns may overlap. Researchers must then separate closely related signals from one another, which becomes increasingly difficult as more labels are added.

In practice, imaging more than three or four targets at once can become an unusually demanding experiment.

NovoTags Were Designed From Scratch

NovoTags are examples of de novo protein design.

Instead of starting with a naturally occurring protein and gradually changing it through repeated laboratory experiments, researchers used computational tools to design entirely new protein structures around selected dye molecules.

The process began with RFdiffusion, an AI system developed by the Baker Lab. RFdiffusion can generate possible three-dimensional protein backbones that fit a requested purpose.

In this case, the researchers wanted small proteins that could surround and bind particular Janelia Fluor dyes.

The dye molecule served as the target. The AI system generated protein structures shaped to hold that target tightly, almost like designing a custom glove around a specific object.

Researchers then used another system, LigandMPNN, to propose amino-acid sequences capable of forming the selected structures.

The resulting candidates were evaluated through programs including AlphaFold and RoseTTAFold before the most promising designs were tested experimentally.

This combination of AI generation, computational filtering, and laboratory validation allowed the researchers to search protein-design possibilities much faster than conventional trial-and-error methods.

The First NovoTags Cover Three Parts of the Visible Spectrum

The researchers developed NovoTags that bind selectively to three Janelia Fluor dyes spanning far-red, orange, and green wavelengths.

These dyes were developed by researchers in Luke Lavis’s laboratory and are known for their brightness, stability, and ability to function inside living cells.

Each designed protein was intended to recognize its matching dye without binding strongly to the others.

That independence is important.

A researcher could attach different NovoTags to different proteins, add the corresponding dyes, and observe multiple cellular structures within the same sample.

The study demonstrated the tags in human cells using advanced fluorescence-imaging techniques.

One published image showed endosomes, mitochondria, and chromatin labeled simultaneously within a HeLa cell. The separate signals allowed researchers to distinguish the cellular transport structures, energy-producing network, and nuclear material in one combined view.

This is only an early demonstration, but it shows how designed tags could help researchers examine relationships among structures rather than studying each one through a separate experiment.

The Technology Could Expand Multicolor Cellular Imaging

Modern fluorescence microscopes can distinguish signals through differences in color, but color is not the only property researchers can measure.

Fluorescent molecules also differ in how long they remain excited before releasing light. This property is known as fluorescence lifetime.

By combining the color of a signal with its fluorescence lifetime, scientists may be able to distinguish labels that would otherwise appear too similar.

The researchers believe combinations of emission spectrum and lifetime information could eventually allow scientists to identify far more tags within one cell.

EMBL reported that future versions of the system could potentially distinguish as many as 30 proteins when spectral separation is combined with fluorescence-lifetime measurements.

That figure represents a future possibility rather than a capability established for routine laboratory use today.

The published research demonstrated three primary NovoTag systems. Reaching dozens of independently identifiable labels will require additional dyes, new protein designs, extensive testing, and reliable methods for separating the resulting signals.

Even so, the potential increase would be significant.

Moving from a handful of visible proteins to dozens could allow scientists to observe complete signaling networks and cellular responses in ways that are currently difficult or impossible.

Researchers Also Created a Molecular Switch Called NovoSplit

The team developed an inducible version of the technology known as NovoSplit.

In this system, a NovoTag is divided into two separate protein pieces. Each half can be attached to a different protein inside the cell.

The two halves remain separate until the matching fluorescent dye is added. The dye then helps bring the pieces together, assembling the complete tag and producing a fluorescent signal.

This makes NovoSplit more than a passive label.

It functions as a chemically controlled molecular switch.

Researchers can decide when the tagged proteins are brought into proximity by controlling when the dye is introduced.

That could help scientists study protein interactions, signaling pathways, and cellular changes that depend on timing.

Instead of only observing that two proteins are near each other, researchers may be able to trigger an interaction and watch what happens next.

The technique could provide a controlled way to examine cause and effect within living cells.

Why Protein Interactions Are So Important

Proteins rarely perform their functions completely alone.

They form complexes, transfer chemical signals, recruit other molecules, and change one another’s activity. Many important cellular processes depend on a specific group of proteins assembling at the correct location and time.

Disease can develop when these interactions go wrong.

Cancer cells may activate growth-related signaling networks that should have been switched off. Neurodegenerative disorders may involve proteins folding incorrectly or forming harmful aggregates. Immune diseases can emerge when signaling pathways become excessively active or fail to respond appropriately.

Researchers often know which proteins are involved but cannot easily see how they interact inside living cells.

A system capable of labeling many proteins at once could reveal the sequence of events rather than providing only isolated snapshots.

Scientists might observe which protein arrives first, how long a complex remains assembled, which cellular compartment it enters, and what changes after a drug or mutation is introduced.

That kind of information could improve understanding of disease mechanisms even before the technology has any direct clinical use.

NovoTags Could Support Longer Imaging Experiments

Fluorescent labels must remain visible without damaging the cell or fading too quickly.

Some fluorescent proteins lose brightness after repeated exposure to light, a process known as photobleaching. Strong illumination can also harm cells, particularly during long experiments.

Janelia Fluor dyes were developed to produce bright, stable signals while functioning effectively inside living systems.

Combining these dyes with small, purpose-built protein binders could allow scientists to track cellular structures over longer periods.

Long-term observation matters because many biological processes unfold over minutes, hours, or days.

Cell division, development, immune responses, organelle movement, and changes caused by disease cannot always be understood through one short imaging session.

More stable labeling could let researchers follow the same cell as it changes, reducing the need to reconstruct a process from separate samples collected at different times.

Smaller Tags May Interfere Less With Cellular Behavior

Adding a fluorescent marker to a protein can affect the protein being studied.

Large tags may alter how a protein folds, where it travels, which molecules it binds, or how quickly the cell breaks it down.

This creates a fundamental challenge. Scientists want the label to reveal normal biological behavior without changing that behavior.

NovoTags were designed as relatively small protein binders.

Their compact size could make them useful in situations where larger fluorescent proteins or enzyme-based tags interfere with the target.

However, smaller does not automatically mean biologically invisible.

Researchers must test whether each tag changes the activity, location, stability, or interactions of the protein to which it is attached.

That validation will be necessary before NovoTags can be trusted across a wide range of cell types and experiments.

The Research Could Connect Light and Electron Microscopy

One of the most ambitious potential uses involves cryo-correlative light and electron microscopy, commonly called cryo-CLEM.

Fluorescence microscopy allows researchers to identify specific molecules by their glowing labels. Electron microscopy provides much greater structural detail but does not always reveal the identity of every structure in the image.

Cryo-CLEM combines the two.

A biological sample is frozen rapidly enough to preserve cellular material close to its natural state. Scientists first locate labeled targets with fluorescence microscopy and then examine the same region with electron microscopy.

The two sets of information can be aligned, allowing researchers to connect molecular identity with highly detailed cellular structure.

The Mahamid Group at EMBL specializes in cryo-electron tomography and related methods for studying molecular assemblies inside cells.

The researchers believe NovoSplit could form the basis of a new generation of tags that are both fluorescent and structurally recognizable within cryo-electron tomograms.

If successful, scientists could identify a particular protein through fluorescence and then examine its surrounding molecular structure at extremely high resolution.

That would help connect what a protein is doing with how it is organized physically inside the cell.

AI Did Not Complete the Research by Itself

The NovoTags project demonstrates the usefulness of AI in biotechnology, but it also shows why experimental science remains essential.

RFdiffusion generated possible protein structures. LigandMPNN proposed amino-acid sequences. AlphaFold and RoseTTAFold helped assess whether the designs were likely to fold as intended.

Those predictions did not prove that the proteins would work.

Researchers still needed to produce candidate proteins and test them through yeast surface display, fluorescence-activated cell sorting, next-generation sequencing, fluorescence-polarization assays, chromatography, structural analysis, and cellular microscopy.

Many computational designs can fail when they are manufactured.

A protein may fold incorrectly, bind too weakly, attach to the wrong molecule, become unstable inside cells, or create unintended biological effects.

AI reduced the search space and helped generate designs that would have been difficult to imagine manually. Laboratory testing determined which designs functioned in reality.

The development is therefore better described as AI-assisted protein engineering than as autonomous scientific discovery.

The Technology Is a Research Tool, Not a Medical Treatment

NovoTags could eventually support disease research and drug development, but they are not medicines.

The proteins were designed to help scientists label and observe molecules in laboratory systems.

They do not repair damaged cells, destroy tumors, restore lost brain function, or deliver therapy to patients.

Their medical value would be indirect.

Better imaging could help researchers identify how a disease begins, which proteins interact, where a drug travels, or how cells respond to treatment.

Those observations could support the discovery of future therapeutic targets or improve the testing of experimental medicines.

The path from a new microscope label to an approved treatment can still take many years.

Scientists must first show that the imaging system is reliable, reproducible, safe for experimental cells, and compatible with different research environments.

NovoTags May Help Researchers Test Medicines More Precisely

Drug development often depends on knowing whether a treatment reaches its intended target and changes the correct biological pathway.

Researchers may measure one protein before and after treatment, but a drug can influence several pathways simultaneously.

A wider multicolor imaging system could reveal those effects within the same cell.

Scientists might watch a medicine bind to one target, alter the location of another protein, activate a repair pathway, and trigger an unintended stress response.

That could make preclinical experiments more informative.

It may also help researchers understand why two cells respond differently to the same drug.

One cell may contain a particular combination of active proteins, while another follows a different signaling route. Observing several molecular events simultaneously could expose those differences.

This remains a potential application rather than a demonstrated clinical benefit.

The Researchers Are Expanding the NovoTag Collection

The first published NovoTags bind three dyes, but the collaboration is already working to develop more.

The project is supported through AI@HHMI, a $500 million Howard Hughes Medical Institute initiative intended to integrate artificial intelligence into scientific research.

The team aims to design tags for approximately a dozen commonly used fluorescent dyes.

Researchers are also interested in binders for dyes that change color or blink on and off. Those properties could support specialized imaging systems and help distinguish more signals within crowded cells.

Future versions may also function as biological sensors.

A NovoTag-based probe might change its fluorescent behavior in response to calcium, metabolites, electrical activity, or another cellular condition.

Instead of merely showing where a protein is located, such a tool could report what is happening around it.

The Designs and Dyes Are Being Shared With Researchers

EMBL reported that the NovoTag sequences and complementary dyes are freely available to the scientific community.

Open availability could accelerate testing because laboratories will not need to recreate the system independently from the beginning.

Researchers can examine how NovoTags perform in different organisms, tissues, microscopes, and cellular environments.

They may also identify weaknesses or develop improvements not anticipated by the original team.

Broad access will be important for determining whether NovoTags become a widely adopted platform or remain useful mainly in specialized imaging laboratories.

Reproducibility is particularly important in biotechnology. A technique must work reliably outside the laboratory that invented it before scientists can depend on it for major conclusions.

Important Limitations Remain

The published results are promising, but the technology is still new.

The researchers demonstrated three principal dye-binding proteins rather than the full collection of dozens envisioned for future imaging.

Many experiments were performed in cultured human cells. Performance may differ in primary cells, tissues, organoids, plants, microorganisms, or living animals.

Delivering the necessary dyes through thick tissues may also be more difficult than introducing them into cells grown in a laboratory dish.

Researchers must determine whether the tags remain stable over long periods and whether attaching them changes the target protein’s function.

As more colors are added, signal separation will become increasingly complicated. Additional labels could also increase the possibility of unintended interactions.

The prediction that researchers may eventually distinguish up to 30 proteins should therefore be treated as a research objective rather than an established capability.

Why This Development Matters

Biology is increasingly moving from studying isolated molecules toward understanding entire systems.

Knowing that one protein exists is useful. Knowing where it moves, which molecules it contacts, when it becomes active, and how the surrounding system responds is much more powerful.

Current imaging tools reveal only a limited part of that activity at one time.

NovoTags introduce a modular approach in which new protein binders can be designed for different dyes rather than relying only on natural proteins or a small set of established tags.

The work also demonstrates a broader change in biotechnology.

Researchers are no longer limited to discovering useful proteins in nature and modifying them gradually. AI systems can help design molecules for particular functions from the beginning.

NovoTags are one example of how that capability may produce scientific instruments, not only medicines.

Key Takeaways

Researchers from the University of Washington’s Institute for Protein Design, Janelia Research Campus, and EMBL Heidelberg developed a new class of fluorescent imaging tools called NovoTags.

NovoTags are small, synthetic proteins designed with AI to bind tightly and selectively to particular Janelia Fluor dyes.

The study demonstrated tags for far-red, orange, and green dyes and used them to label cellular structures in human cells.

The researchers also created NovoSplit, an inducible system in which a dye brings two protein-tag fragments together and allows scientists to control protein interactions.

Future versions could expand multicolor imaging and potentially allow researchers to distinguish many more proteins through combinations of color and fluorescence lifetime.

The system may eventually help connect fluorescence microscopy with cryo-electron tomography, giving researchers both molecular identification and detailed structural information.

NovoTags are laboratory research tools. They are not medical treatments, diagnostic products, or proof that AI can independently conduct biotechnology research.

Frequently Asked Questions

What Are NovoTags?

NovoTags are small proteins designed from scratch with AI to bind specific fluorescent dyes. Scientists can attach them to proteins or cellular structures they want to observe.

Were the Proteins Found in Nature?

No. They were created through de novo protein design, meaning their structures and sequences were computationally designed rather than copied directly from naturally occurring proteins.

Which AI System Was Used?

Researchers used RFdiffusion to design protein backbones and LigandMPNN to generate amino-acid sequences. AlphaFold and RoseTTAFold helped filter the designs before laboratory testing.

How Many NovoTags Were Demonstrated?

The study developed tags for three Janelia Fluor dyes spanning far-red, orange, and green wavelengths.

Can Scientists See 30 Proteins at Once With NovoTags Now?

Not as a routine demonstrated capability. Researchers believe future combinations of color and fluorescence-lifetime information could potentially distinguish up to 30 labels, but considerably more development and validation are required.

What Is NovoSplit?

NovoSplit divides a designed tag into two pieces attached to different proteins. Adding the matching dye brings the pieces together, allowing researchers to trigger and observe a controlled protein interaction.

Could NovoTags Help Study Disease?

Potentially. They may help researchers examine protein interactions, drug responses, cellular structures, and disease pathways. They are not treatments or clinical diagnostic tests.

Are NovoTags Available to Other Researchers?

EMBL reports that the tag sequences and complementary dyes are being made freely available to the scientific community.

When Was the Research Published?

The peer-reviewed paper was published in Science on July 16, 2026. EMBL announced and explained the findings on July 17.

Support New To Education

New To Education provides independent reporting and accessible analysis covering biotechnology, artificial intelligence, health research, science, engineering, education, and the discoveries shaping future careers and industries.

Readers can support our work by sharing this article, exploring our educational and professional services, joining the New To Education community, or contributing through the support options available on our website.

Every visit, share, booking, and contribution helps us continue translating complex scientific research into useful information for students, educators, professionals, families, and lifelong learners.

Peking University Expands “AI for Science” as China Reshapes Higher Education Around National Innovation

https://www.newtoeducation.com/view-blog/peking-university-expands-ai-for-science-as-china-reshapes-higher-education-around-national-innovation-6a585f690b47d

Mayo Clinic Develops AI Tool That May Detect Pancreatic Cancer Years Earlier

https://newtoeducation.com/view-blog/mayo-clinic-develops-ai-tool-that-may-detect-pancreatic-cancer-years-earlier-6a45f35c6bbd8

Final Thoughts

NovoTags reveal a side of artificial intelligence that receives less attention than chatbots, image generators, and consumer software.

AI is beginning to help scientists design physical biological tools that have never existed in nature.

In this case, researchers used computational systems to build small proteins shaped around specific fluorescent dyes. Those proteins can be attached to molecules inside living cells, turning otherwise invisible biological activity into signals that scientists can observe.

The achievement is not simply that the tags glow.

Scientists have had fluorescent labels for decades.

The larger advance is the possibility of designing many independent labeling systems for different colors, structures, and experimental needs.

That could allow researchers to move beyond observing one or two proteins and begin watching entire groups of molecules communicate within the same living cell.

NovoSplit adds another layer by turning the fluorescent system into a controllable molecular switch. Researchers may be able to trigger a protein interaction and then observe the resulting cellular response.

The eventual connection with cryo-electron tomography could be even more powerful. Scientists could first identify a protein through fluorescence and then examine its structural surroundings at extremely high resolution.

The technology remains at an early stage. Three tags are not yet a complete cellular-imaging language, and the possibility of distinguishing dozens of proteins still needs to be proven through additional experiments.

Researchers must also show that the tags do not interfere significantly with the proteins they are intended to observe.

Still, NovoTags point toward an important future for biotechnology.

Scientists may increasingly design research tools according to the biological questions they want to answer rather than accepting the limited tools evolution happens to provide.

The cell has always contained far more activity than researchers could see.

AI-designed proteins may help turn on a few more lights.