AI-designed enzymes

Welcome to Plenty of Room!

Today we are going back into enzyme design, with a big AI twist! Is this an unexpected series?

Like what you read? Share it with someone who would too!

Plenty of Room is your guide to the cutting-edge news related to molecular machines.

Already subscribed? Share with a friend that might find this interesting! It really helps.

New here? Just go ahead and subscribe!

Let’s get into it now.

AI-designed enzymes

We were talking about enzymes just like week, here on Plenty of Room: biological catalysts which speed up biological chemical reactions, allowing life to exist. These powerful proteins are also important in sustainable chemistry, drug production and biotechnology, so there is a lot of interest in designing new ones, to create new materials and to get rid of persistent pollutants (looking at you, plastics and PFAS)! And there has been a lot of work in this direction, especially when it comes to computational design. Unfortunately, the field has long struggled with some key limitations:

• Rigid backbone constraints: Previous designs relied heavily on existing protein scaffolds, limiting flexibility.

• Poor Active Site Preorganization: Getting catalytic residues precisely aligned for multi-step reactions is incredibly tricky.

• Low Efficiency: Many designed enzymes struggled with catalytic turnover, making them slow and ineffective.

So, while previous de-novo design of enzyme was achieved, they were limited to simple reactions, with only one or two reaction steps. The more steps there are in a reaction, the harder is to design an enzyme, because the catalytic action is based on a sequence of structural shifts: the enzyme has to guide the substrate into place, stabilize intermediate states, and then release the products.

All of this brings us to today’s paper, where the researchers developed a new de novo enzyme design pipeline, based on not one, but two AI-based tools! Their goal was to recreate a serine hydrolase, a versatile enzyme class involved in processes from digestion to plastic degradation! In addition, they are also very complicated, with natural enzymes using a six step process to break ester bonds. So, what did they use? They majorly leveraged 2 computational tools:

• RFdiffusion: do you remember this? RFdiffusion can design novel protein backbones that, in this case, support catalytic sites. We have seen its power multiple times lately, it’s really shaping up to be a versatile, staple tool!

• PLACER (Protein-Ligand Atomistic Conformational Ensemble Resolver): the researchers developed this deep neural network that evaluates the compatibility of designed active sites with each step of the reaction mechanism. Very cool! It was the base for their filtering process.

By combining these tools, the authors showed how enzyme preorganization is a crucial step for efficient catalysis, and then managed to create new designs with unprecedented accuracy! The researchers followed a multi-step design-build-test cycle:

1. Scaffold Generation with RFdiffusion

   RFdiffusion was used to generate new protein backbones from scratch that conform to the required active site geometry​, overcoming the limitations of fixed scaffolds

2. Active Site Optimization with PLACER

   Enzyme catalysis is a multi-step process, requiring stabilization of multiple intermediates. The researchers trained PLACER on thousands of protein-small molecule complexes and used to evaluate each designed enzyme for its ability to stabilize transition states throughout the catalytic cycle. By filtering designs through PLACER, only the most catalytically competent enzymes were selected for testing.

3. Sequence Design and Refinement

   LigandMPNN, a deep learning model for sequence optimization, was used to design amino acid sequences that stabilize the desired backbone structures and Rosetta FastRelax was applied to refine the sequences while maintaining the catalytic geometry

After this pipeline, they had a first, quick filtering step with AlphaFold2, to predict the structural stability of the proteins. In a second step, they used PLACER to evaluate whether the active sites were properly preorganized for catalysis, using a simplified active site. Designs that passed this filter were expressed in E. coli and tested the activity of the enzymes using fluorescent substrates. After round 2, the researchers had 5.2% active enzymes, compared to only 1.6% in round 1, showing that the filtering with PLACER increases the number of functional proteins. Building on this, they added a third round of filtering, where PLACER was used to model five critical reaction states for the catalysis, evaluating the entire reaction! With this addition, 18% of designs showed activity, and two exhibited catalytic turnover, meaning that they could complete the reaction cycle without getting stuck, a major milestone! And the researchers managed to even improve these designs, leading to a crazy 60000x improvement over previously designed esterases! They also used crystallography to confirm at atomic resolution the accuracy of their designs.

This work is a giant step forward in enzyme engineering. While the enzyme that they designed are a lot slower than their natural counterpart, they are in the range for natural enzymes, and no one before had ever done something like this! And while the process requires less lab work than traditional enzyme screenings, it is still quite complex and require a lot of expert intuition to figure out the best direction. But it is still a very impressive work! Go and read it for yourself.

And as always, thank you for reading!

​

In other news:

• New drugs on the block: Oligonucleotide drugs face delivery challenges due to short half-life and poor uptake. Nanobodies, small antibody fragments, could offer a promising solution with high affinity, better tissue penetration, and easy production. This review explores NucleoBodies, nanobody-oligonucleotide conjugates, for targeted delivery, diagnostics, and gene therapy, paving the way for precision medicine.

• More ways for icosahedral particles: If you like icosahedral structures, like viruses or nanoparticles, this article might be for you. It presents a universal framework for designing multi-component icosahedral structures using concentric shells of chiral and achiral particles. By mapping shell sequences onto a hexagonal lattice, the authors establish simple design rules and optimize size mismatches for stable assembly. Crazy but interesting!

• Sensing with DNA origami: Electrochemical sensing, in particular. This review explores the potential of DNA origami for assembling nanostructures with precisely positioned sensing elements, enhancing the organization and functionality of biosensors. While its adoption in electrochemistry is still limited, growing research efforts could unlock next-generation multifunctional electrochemical sensors. It’s a very good review, don’t miss it!

• Not yet a subscriber to Plenty of Room? Sign up today — it’s free!

• You think a friend or a colleague might enjoy reading this? Don’t hesitate to share it with them!

• Have a tip or story idea you want to share? Email me — I’d love to hear from you!

• You have something you would love me to cover? Just reach out here or on my social!