AI protein design goes big

Welcome to Plenty of Room!

Today, we are back with news about computational protein design! And I am guessing that we will see even more moving forward, with the Nobel prizes and all that.

Plenty of Room is your guide to the cutting-edge news related to molecular machines. New here? Just go ahead and subscribe! Already subscribed? Help a friend or a colleague save some time and share this with them!

Let’s get into it now.

AI protein design goes big

Designing proteins with specific structures and functions is a base for advances in synthetic biology, medicine, and biotechnology. Being able to create custom proteins can unlock solutions for problems in drug development, industrial catalysis, biosensors, and more! Unfortunately, protein engineering is hard, especially for large proteins, and traditional methods often get “stuck” in inefficient paths. To refine these structures experimentally, scientists sometimes need to test hundreds of variants, a process that’s slow and costly, as I covered another time.

In today’s paper, the team developed Relaxed Sequence Optimization (RSO), a new computational pipeline that improves structural prediction and design accuracy by working within a continuous sequence space. So, what is the difference? Traditional algorithms define rigid amino acid assignments at each step: this creates problems for the algorithm, making the whole process inefficient. Instead, RSO uses “floating” intermediate amino acid states, creating pseudo-sequences for optimized 3D shapes without forcing each position into a specific amino acid right away. This smooths out the process and lets RSO explore more potential configurations, ultimately finding better solutions.

In practice, the RSO pipeline integrates AlphaFold2 (AF2) to predict 3D structures of candidate sequences. AF2 is used to compute structural deviations (measured by root-mean-square deviation or RMSD) from the target structure, which allows RSO to use this loss gradient to guide sequence adjustments iteratively. Once RSO converges on a relaxed sequence and 3D structure, they are passed to ProteinMPNN, a neural network-based tool that creates sequences that fold into a specific 3D structure. ProteinMPNN ensures that the final sequence folds correctly and can actually be synthesized. In the last step, AF2 checks the optimized sequences, verifying their structural match with the original design..

To test RSO, the team designed over 100 proteins with various structures, including large single-chain proteins (over 1000 amino acids!), protein binders, and complex scaffolds. RSO excelled at achieving low RMSD values (a measure of structural accuracy), especially for larger proteins where traditional methods tend to fail. The researchers then took things a step further, synthesizing several proteins and confirming their structures with X-ray crystallography and cryo-EM. The strong match between the predicted and real-world structures underscores RSO’s precision.

So, RSO represents a big advance in scalable and precise protein design. Its flexibility, coupled with high accuracy, offers promising applications in different fields:

• Drug discovery: RSO’s ability to accurately model large proteins could lead to therapeutic proteins with custom structures and functions. Its ability to handle large proteins and complex folding requirements is interesting because it starts approaching the size of antibodies, and creating completely custom antibody would be a huge achievement.

• Synthetic biology: The flexibility of RSO makes it ideal for synthetic biology applications, from biocatalysts that drive green chemistry to highly specialized biosensors..

But honestly, there is a lot more in the original paper, so head here and read it for yourself!

In other news:

• Nanotech for climate change: If you are interested in the role nanotech can have in fighting climate change, this comment article is for you. A diverse group of nanoscientists from various sectors identifies four key areas where nanotechnology can drive progress: energy storage (batteries), catalysis, coatings and interface technology, and greenhouse gas capture. Very inspiring!

• Folding proteins in collaboration: If you are interested in computational protein design, but it’s all a bit too intimidating (just like I feel), you want to look at this paper. This article introduces ColabFold-AF2, an open-source tool that simplifies using AlphaFold2 (AF2) for predicting protein structures. ColabFold-AF2 runs on Google Colaboratory and includes a command-line tool, offering quick turnaround times and optimized access to AF2’s models. The protocol is accessible to both beginners and advanced users, with typical procedures taking less than 2 hours.

• Shaping nanocrystals: For some more traditional nanotech, this study investigates how the shape of ZnS nanocrystals (NCs) affects the surface density of oleylamine (OLA) ligands. These findings provide insights into ligand arrangement on NCs, which could help in fine-tuning NC morphology and chemical functionality for specific applications, such as in cameras and sensors.

• 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!