AI-Designed Proteins Revolutionize Fluorescent Cell Imaging

Janelia Senior Group Leader Luke Lavis is renowned for his development of exceptionally bright fluorescent dyes. Meanwhile, HHMI Investigator David Baker is known for crafting proteins that do not exist in nature.

So, what unfolds when you combine these two innovative minds?

This collaboration is part of the AI@HHMIexternal link, opens in a new tab initiative, aimed at producing advanced tools that will enable biologists to observe cellular structures and processes with unmatched clarity.

Leveraging an AI model created by the Baker Lab, the team is designing new proteins that attach to specific Janelia Fluor dyes developed by the Lavis Labexternal link, opens in a new tab. This results in the creation of a novel fluorescent probe known as NovoTag.

The integration of Baker’s cutting-edge AI technology for generating small protein binders and Lavis’s innovative, photostable, and cell-permeable Janelia Fluor (JF) dyes is anticipated to facilitate biological imaging experiments that are otherwise not feasible with existing equipment.

“The AI@HHMI project enables David’s lab to advance the frontier of de novo protein design, while my lab contributes next-generation dyes. Together, we’re creating a robust and validated toolkit,” Lavis explains. “Our goal is to establish labeling systems that will enhance or replace the current methods.”

Challenges in Cellular Imaging

While fluorescent dyes are excellent for illuminating cellular structures, they require a molecular tag—like HaloTag or SNAP-tag—to specifically image proteins.

These systems have facilitated numerous biological discoveries. However, since they each rely on the same chemical connector, researchers are limited to using just one or two colors, thereby restricting their ability to observe multiple proteins simultaneously. If scientists wish to explore intricate signaling pathways involving several proteins, they must conduct multiple experiments, each time imaging only a few proteins, before synthesizing the collected data.

This limitation hinders a comprehensive understanding of cellular processes and the interactions among proteins that enable various functions.

“Currently, conducting experiments with more than three or four colors is a significant challenge,” Lavis notes. “The difficulties extend beyond imaging to devising effective labeling strategies.”

Creating AI-Enhanced Proteins

The AI@HHMI project aims to address these challenges by utilizing AI to generate NovoTags—a collection of small proteins that can directly bind to the various JF dyes, creating a platform for labeling without the need for conventional chemical connectors. By engineering distinct NovoTags for each JF dye, researchers will be able to visualize numerous colors and proteins simultaneously.

The RFdiffusionexternal link, opens in a new tab model, developed by the Baker Lab, allows the design of new proteins that do not exist naturally and can bind effectively to small molecules like JF dyes. Unlike natural proteins—which often require extensive modification to fit specific dyes—RFdiffusion enables the creation of proteins that are optimized for binding to particular JF dyes from the outset.

This AI tool predicts how a new protein will fold and attach to the dye, generating thousands of potential designs that help researchers identify optimal candidates for lab testing, explains Long Tran, a graduate student in the Baker Lab who co-led recent researchexternal link, opens in a new tab forming the foundation of the AI@HHMI initiative.

“Designing a three-dimensional protein structure that can encapsulate a small molecule is something that AI makes possible,” Lavis adds.

A New Era of Probes

The team has recently demonstrated their ability to generate three distinct proteins, each selectively binding to a different JF dye color. They collaborated with Julia Mahamid’s team at EMBL-Heidelberg to validate these systems in sophisticated microscopy experiments. Upcoming plans include developing a series of NovoTag probes for approximately a dozen different color dyes commonly used in microscopy, enabling scientists to possibly image multiple proteins concurrently.

Additionally, the team is looking to innovate further with advanced tools, including binders for dyes that can change colors or blink on and off. They are also applying Janelia’s expertise in designing fluorescent indicators to create new probes capable of measuring specific physiological signals, such as calcium levels or metabolites.

The team intends to make NovoTags accessible to the entire scientific community.

“This imaging toolkit will allow researchers to capture several organelles, proteins, and cell compartments simultaneously,” states Tran. “It represents a remarkable advancement over what HaloTag, SNAP-tag, and traditional fluorescent proteins can currently achieve.”