Skip to main content

Weekly Reading: Five tasks that still challenge protein designers (Nature 635, 246-248 (2024))

Original article is here ( Five protein-design questions that still challenge AI)

This article brings an insightful snapshot for anyone interested in protein design, expecially non-expert (such as me!) consideing starting their own mini-project in this field. As an enzymologist, one of the striking aspects of the article is the emphasis on protein conformation. We often overlook the fact that proteins and enzymes are not static - they are indeed dynamic molecules. The structures we see are merely snapshots of one moment in time, but they does not necessarily capture the full extent of the protein's motion. 

If protein dynamics are a 4-dimensional problem, perhaps designers are only starting to scratch the surface, seeking to understand this complexity in a simplified form. (I mean how many enzyme people run MD by themselves?) This raises a question: can we truly extract or capture protein dynamics just from sequence information alone? This seems like a huge challenge, expecially when considering how much motion, dynamics, and flexibility contribute to a protein/enzyme's function.

The article also made me relfect on the role of traditional wet-lab scientists in this rapidly evolving field. As AlQuaraishi mentioned, better nad more data will lead to better protein designs. So, will wet-lab scientists only serve as data gatherers for computational models, providing the missing pieces? What is the future of wet-lab haha 

Another aspect the article doesn’t fully address is the (in silico and in vivo) screening process for protein designs. While generative AI tools like AlphaFold are invaluable for predicting protein folding, how do we screen the designs in a way that ensures they will perform as intended? Developing general, high-throughput assays has been remained a significant challenge in the field. Success of directed evolution most of time relies on the successful desiging of screening platform. If we could develop more in-silico methods for screening designs—beyond simply assessing foldability—we might take a major leap forward in the protein engineering process.

Finally, I’m left wondering about the models used in protein design. Most of the algorithms driving these tools seem to be based on large language models (I think—correct me if I’m wrong!). The underlying architecture of these algorithms—how it plays a role in protein design—is still something I can’t fully answer as a wet-lab scientist. But it adds another layer of complexity to the entire design process, making it even more fascinating (and daunting).

Labels:

Comments

Post a Comment

Popular posts from this blog

Primer : Understanding PDB vs CIF files - what's inside your pymol file?

Whenever you visualize a protein structure using PyMOL or Chimera , have you ever wondered what’s really inside the file ? Nowadays, I am getting into informatics, and hope that this series helps beginners like me to get into it. Today, let's explore the PDB and CIF file formats , and their differences, Let's start by going to RCSB PDB and searching for 5GZA , a kinase enzyme that interacts with mannose sugar. Click the download tab and select PDB and CIF file formats. Exploring the PDB File Format To visualize the contents of a PDB file , open it in VS Code (or any text editor—though I personally dislike Notepad!). PDB files contain structured information with well-defined column formats . The first few sections include: HEADER, TITLE, COMPND → Metadata about the protein SOURCE, AUTHOR → Experimental details ATOM & HETATM → Atomic coordinates and ligands CONECT → Connectivity (bonds between atoms) The most useful part is the ATOM section , which contains 3D atomic ...
Post a Comment
Read more

Molecule 101: Urea (1)

U rea, this small molecule has been a lifesaver in my lab work, especially when dealing with stubborn western blots for transmembrane proteins. Urea, composed of two amide groups (Figure 1), i s one of the most widely used protein denaturants in protein biochemistry.   Fig 1. Chemical structure of Urea (Fun fact: if you’re looking for smoother hands, check out hand creams containing urea—it’s also a moisturizing agent!) One of urea’s favorite applications is in proteomics, where it is used to unfold and solubilize proteomes, making them susceptible to trypsin digestion. Its history in protein science goes back to Anfinsen’s famous experiment, which demonstrated that the code for protein folding reside in the amino acid sequence. Anfinsen unfolded ribonuclease using urea, and upon removing it, some proteins refolded, proving the “code” for folding is intrinsic to the sequence. Like many scientists, I’ve been using urea in my experiments for years without giving much thought to how i...
Post a Comment
Read more