AI@HHMI: Lighting up Life inside cells with AI-designed proteins

Key Takeaways:

• A team led by HHMI Investigator David Baker and Janelia Senior Group Leader Luke Lavis is using AI to create a new class of fluorescent imaging probes called NovoTags.
• The team is using an AI model developed by the Baker Lab to create novel, small proteins that bind to fluorescent dyes developed by the Lavis Lab, creating a suite of new probes for detecting proteins and structures inside cells.
• These next-generation tools will allow researchers worldwide to see many proteins at once over longer time periods — experiments that can’t be done with current probes, potentially helping to accelerate scientific discovery.
• The project is part of AI@HHMI, the institute’s $500 million initiative to support AI-driven projects and embed AI systems throughout the scientific process. 

Janelia Senior Group Leader Luke Lavis is famous for developing fluorescent dyes brighter than anything seen before. HHMI Investigator David Baker is renowned for creating proteins not found in nature.

So, what happens when you combine the two?

An AI@HHMI project to create new tools that will allow biologists to see the structures and processes happening inside cells with unprecedented clarity.

Using an AI model developed by the Baker Lab, the team is creating novel proteins that bind to specific Janelia Fluor dyes pioneered by the Lavis Lab, producing a new type of fluorescent probe they call NovoTag.

The team hopes that the combination of Baker’s groundbreaking AI technology for creating small protein binders from scratch and Lavis’s bright, photostable, and cell-permeable Janelia Fluor (JF) dyes will enable new biological imaging experiments that can’t be done with the currently available systems.

“The AI@HHMI project allows David’s lab to push the frontier of de novo protein design, my lab to make next-generation dyes and then combine them in this project to make a robust and validated set of tools,” Lavis says. “We aim to develop labeling systems that are going to complement or replace existing ones.”

Limits To What Biologists Can See

Fluorescent dyes are great for illuminating structures inside cells, but to image specific proteins, they need to be paired with a molecular tag — like HaloTag or SNAP-tag — that’s attached to the proteins the scientists want to study.

These systems have enabled many biological discoveries, but because they each rely on the same chemical connector, researchers can only use one or two colors  — and therefore only see one or two proteins — at a time. If they want to investigate a complicated signaling pathway involving many different proteins, they need to image a few proteins at a time over multiple experiments and then piece the information together.

That limits biologists from truly seeing what is happening inside cells and how all the proteins interact to carry out their functions.

“Right now, it’s a heroic experiment to do more than three or four colors,” Lavis says. “The imaging is hard but also just figuring out the labeling strategy is hard.”

Designing AI-Powered Proteins

The AI@HHMI project aims to overcome this problem by employing AI to create NovoTags — a series of novel, small proteins that bind directly to different JF dyes, providing a platform for labeling without the requirement of traditional chemical connector. By creating independent NovoTags for each color JF dye, scientists can image many colors — and proteins — simultaneously.

RFdiffusion, an AI model developed by the Baker Lab, enables the team to build new proteins not found in nature that bind to small molecules — in this case the JF dyes. Unlike naturally occurring proteins that need to be modified through many rounds of engineering to fit with a particular dye, RFdiffusion allows scientists to create a protein that tightly binds to a specific JF dye from the get-go, acting like a perfect catcher’s mitt for a particular dye molecule.

The AI-powered tool predicts how a new protein will fold and bind to the dye and also generates thousands of possible designs to help researchers home in on which ones to create in the lab, says Long Tran, a graduate student in the Baker Lab who co-led recent research that forms the basis of the new AI@HHMI project.

“Building a three-dimensional protein that surrounds a small molecule is just something you can’t do without AI,” Lavis says.

A Suite of New Probes

The team recently showed that they can generate three unique proteins that selectively bind three different color JF dyes. They then worked with Julia Mahamid’s team at EMBL-Heidelberg to test these systems in advanced microscopy experiments. Next, they plan to develop a series of NovoTag probes for about a dozen different color dyes commonly used in microscopy, allowing scientists to potentially image many proteins at once.

In addition, the team plans to expand this work to create more advanced tools, including binders for dyes that change color or blink on and off. They are also leveraging Janelia’s expertise in fluorescent indicator design to make new probes that can measure specific physiological signals like calcium or metabolites.

They plan to make the NovoTags available to the entire scientific community.

“You can have this great imaging toolkit that could image several organelles, several proteins, several compartments, at once,” Tran says. “It will really capture a lot of biological interactions that HaloTag and SNAP-Tag and fluorescent proteins are currently not able to do.”