AI-Designed Protein Nanoassembly: Molecular Lego's Next Frontier!

Creating modular nanostructures? Until now, it has been the domain of DNA origami. Is it changing? Maybe! Today, we jump into how new protein materials can be created from a set of modular, LEGO-like building blocks!

Proteins Play LEGO

Proteins are the functional building blocks of life, fueling practically everything. You know that.

In nature, they don’t work alone. Proteins interact with each other to create complex structures, transmit signals, or perform difficult chemical reactions. These interactions follow a simple and modular set of binding rules, and bonds can be combined endlessly.

This simple system inspired researchers to explore ways to assemble molecules. My favourite, of course, is DNA base-pairing, which can create complex structures, like in DNA origami. And there are other ways, like metal coordination or host-guest chemical interactions. We are pretty good at it!

But designing protein assemblies is still hard. When you want two proteins to interact, you need to optimize the sequences on both sides and keep them foldable. A much more complex system than simple DNA!

Why get into all this trouble? Rationally designed protein nanomaterials could be amazing for drug delivery, catalysis, biosensing, synthetic cells, and more! Think DNA origami, plus the amazing functions of proteins.

The main problem is control: reliably pushing proteins to self-assemble in designed structures, instead of messy aggregates.

The Modular Idea

This is where this week’s paper comes in. The team developed a whole new way to design assemblies formed by multiple proteins!

Previous attempts designed a target shape with building blocks, and then connected them with protein interfaces. Here, the team flipped the approach. They place the bonding interfaces in space first, and then create the structural blocks to form a shape!

The system has two parts (both proteins, just to be clear):

• Building blocks: Protein oligomers that form the shape, like edges and vertices in a polyhedron.

• Bonding modules: Designed interfaces that act as directional, programmable bonds to connect two building blocks. The bonding modules are based on a set of proteins called LHDs.

The building blocks are reusable, and you can mix-and-match blocks and bonds to build different architectures using the same “part list”!

Computational Pipeline: 3 Steps to Complete Assemblies

The computational pipeline has three main steps:

1. Define the architecture & choose modules. You pick the structural building blocks and the bonding modules, and arrange them in space to have the required angles for your structure.

2. Bridge gaps with rigid junctions and backbones. Computationally create rigid junction adapters to bridge and fix the orientations of the building blocks and binding modules. The team used two approaches: template fusion using WORMS (an algorithm good for extended/pyramidal geometries) or RFdiffusion (a generative deep network that produces compact, rigid junction backbones, better for 2D arrays and compact assemblies).

3. Sequence design & in-silico filtering. The sequences for the junctions are designed using ProteinMPNN (an algorithm optimizing sequences from the protein structure) and then filtered using AlphaFold2, to select candidates predicted to form as wanted!

Results: A Zoo of Protein Assemblies!

After design, all the proteins were expressed in E. coli, screened with SEC, and characterized using negative staining EM and cryo-EM.

Here comes the exciting part: let’s look at what they built!

Two-component polyhedral cages

They started with a simpler system, designing 64 (!) two-component cages with different geometries. They used only 12 types of building blocks and one binding module!

37 passed the first experimental checks, with 18 easily purified. The team confirmed the formation of 11 of the cages using negative staining EM and 2 using cryo-EM! The high-resolution reconstructions of the nanocages matched the models very closely.

Interacting networks with reusable building blocks

They showed how one building block can produce assemblies with different partners.

In one example, a component produced five nanocages, with dimeric, tetrameric, icosahedral, and octahedral variants. This highlighted the power of the shareable interfaces!

Three-component assemblies

Once they mastered the “easy” stuff, they moved to a harder target: extend the method to three-component systems.

Here, the team experimented with 2 distinct interfaces instead of a single interaction surface. This is like using 2 types of bonds rather than just one to build more complex structures.

The team built pyramidal and dihedral assemblies. Of the 24 tested three-component designs, 6 reconstructions matched well to the design. The team also noticed that the success rates were higher when at least one component was previously validated (~30–50%) vs. both new (~10–20%).

2D (Dynamic) Lattices

The next step is to design reconfigurable 2D lattices. In this case, small deviations from the design can create accumulated strain that will destroy the design!

The researchers used a two-component system, with one of them already validated and the other one new. Out of 6 designs, 3 could be purified, and 2 actually created a 2D lattice!

The team also studied the dynamic reconfiguration. Preformed 2D layers could be dissolved and reconfigured into closed cages by adding a competing component!

3D Crystals and Hierarchical Assemblies

To finish it off, they designed 3D crystals. They used octahedral cages with outward bonding modules, and then designed linkers to connect the cages into cubic lattices.

Mixing the cages with the linker produced polycrystalline assemblies, with EM views matching the designed lattice projections! The crystals had relatively small domain sizes, probably because of the strong bonding interactions.

A Worthy Opponent for DNA Origami?

Cool work! Even if a bit hard to read.

The team here wanted to create a simpler way to create protein assemblies. And they delivered! I would love to see more applications of similar technologies. I know many therapeutics startups are using de novo protein design, but I would love to see something different!

The big limitation of this method is the variability in the success rate, because complex, multicomponent, or fully novel designs have lower yields. And of course, they can aggregate, but that happens!

I see DNA origami as the standard for bio-nano-materials. And proteins are still far behind, even with these advances. But they are slowly getting there! And proteins have incredible functions, and the same can’t be said of the inert DNA. Plus, proteins can be more easily scaled up than DNA origami!

So, cool work, a bit hard to read. But great images! Go and read it here.

If you made it this far, thank you! Do you see a future for de novo protein design? Do you use it? What do you think could be the next big target? Reply and let me know!

More Room:

• Throwing SHADEs at DNA: Interested in sensing pH with DNA? This paper presents SHADE (shadow-strand hybridization-actuated displacement engineering), a DNA-based strategy for precise and programmable pH sensing. By using shadow strands to control i-motif folding and hairpin structures for alkaline response, SHADE enables highly tunable pH detection. When combined with aptamers, it creates cell-targeted probes that produce strong fluorescence in acidic tumor environments, improving in vivo imaging. SHADE’s programmability also offers potential for molecular devices and signal transduction applications.

• Studying NanoStuff in DNA NanoPores: Super cool work! This study investigates how nanoconfinement affects molecular interaction kinetics using DNA origami nanopores. The researchers placed single receptors inside or outside narrow pores and tested binding with differently sized ligands. They found that nanoconfinement slows both ligand association and dissociation, even below the diffusion limit, but these effects balance out, resulting in no change in overall binding affinity. Additionally, ligand trapping and local concentration enhancement were not observed. The findings provide insights for designing biosensing nanopores and understanding biological nanochannels.

• More Uses For CRISPR: Did you think CRISPR-Cas was only for gene editing? Think again. This study introduces cc-LFA (Cas12a cis-cleavage mediated lateral flow assay), a highly specific CRISPR-based diagnostic platform. cc-LFA uses a double-key recognition mechanism combining Cas12a cis-cleavage with invasive hybridization. Integrated with multiplexed nucleic acid amplification and lateral flow detection, cc-LFA achieves single-base resolution, >90% sensitivity, and 100% specificity in detecting multiple respiratory pathogens and nine high-risk HPV subtypes. A portable automated device further demonstrates its potential for decentralized, point-of-care diagnostics.

• Share Plenty of Room with founders or builders

  I help biotech and deep tech companies transform complex technologies into engaging content that builds credibility with investors, partners, and potential hires. Let’s chat!

• Know someone who’d love this?
  Pass it on! Sharing is the easiest way to support the newsletter and spark new ideas in your circle.

• Got a tip, paper, or topic you want me to cover?
  I’d love to hear from you! Just reply to this email or reach out on social.