Enzymes are the engine of life. Machine learning helps scientists design new things

Enzymes are molecular machines that carry out chemical reactions that sustain all life, and are the ability to attract the attention of scientists like me.

Consider the movement of your muscles. Your body releases a molecule called acetylcholine, causing contraction in muscle cells. Acetylcholine sticks too long, and can paralyze muscles, including cardiomyocytes. This is where the enzyme acetylcholinesterase appears. This enzyme can break down thousands of acetylcholine molecules per second, avoiding paralysis, and continuing to live. Without this enzyme, it would take a month for the acetylcholine molecule to break down on its own.

I can imagine why enzymes are particularly interesting for scientists trying to solve modern problems. What if there was a way to destroy cancer cells as quickly as acetylcholinesterase breaks down acetylcholine? If the world needs to act quickly, enzymes are a compelling candidate for work.

Unfortunately, enzyme design is extremely difficult. It’s like working with an atom size Lego set, but the instructions are lost and the things will not be held together unless they are fully assembled. A newly published study from our team suggests that machine learning can act as architects for this LEGO set, helping scientists to accurately construct these complex molecular structures.

What is an enzyme?

Let’s take a closer look at what makes up enzymes.

Enzymes are proteins. This is a molecule that enlarges the molecules that function behind the scenes that keep all living things alive. These proteins are made up of amino acids. Amino acids consist of a set of building blocks that can be sewn together to form long strings that are tied to a particular shape.

The specific structure of a protein is the key to its function, just like the shape of everyday objects. For example, a spoon is designed to be designed to hold liquids in ways that a knife simply cannot. Enzymes involved in muscle movement are not very suitable for plant photosynthesis.

For the enzyme to function, it adopts a shape that perfectly matches the molecules of the process, so that the lock matches the key. The unique groove of the enzyme – lock – interacts with the target molecule – keys are found in the region of the enzyme known as the active site.

The active sites of the enzymes precisely release amino acids to interact with the target molecule when it enters. This makes it easier for molecules to undergo chemical reactions and turn into something else, which speeds up the process. After a chemical reaction occurs, a new molecule is released and the enzyme is ready to process another molecule.

How do you design enzymes?

Scientists have spent decades trying to design their own enzymes to create new molecules, materials, or therapeutics. However, it is extremely difficult to make enzymes that look as fast as they are naturally found.

Enzymes have complex, irregular shapes made up of hundreds of amino acids. Each of these building blocks must be fully placed. Otherwise, the enzyme will slow or stop completely. The difference between a speed racer and a slow-pork enzyme can be less distance than the width of a single atom.

Initially, scientists focused on modifying the amino acid sequence of existing enzymes to improve speed or stability. Early success with this approach was primarily able to improve enzyme stability and catalyze chemical reactions at a higher range of temperatures. However, this approach did not help much in improving the speed of the enzyme. To date, designing new enzymes by modifying individual amino acids is generally not an effective way to improve natural enzymes.

Researchers have found that using a process called induced evolution proves to be much more fruitful, so that the amino acid sequence of an enzyme changes randomly until it is able to perform the desired function. For example, studies have shown that directed evolution can improve chemical reaction rates, thermal stability, and even produce enzymes with properties that are not found in their properties. However, this approach is usually labor-intensive. You need to screen many mutants and find out what you need to do. In some cases, this method may not work at all if there is no suitable enzyme to start.

Both of these approaches are limited by their dependence on natural enzymes. This means that limiting your design to the shape of natural proteins can limit the types of chemistry that enzymes can promote. Remember, you cannot eat soup with a knife.

Is it possible to make enzymes from scratch instead of changing natural recipes? Yes, on the computer.

Design enzymes on a computer

The first attempt to calculate enzymes relies heavily on natural enzymes as the starting point, focusing on placing the enzyme active site on the native protein.

This approach is similar to finding suits at a thrift store. The geometry of the active site of the enzyme (the body of this analogy) is very specific and is unlikely to fit perfectly, so rigid random protein fixation structures (suits with random measurements) are completely It is unlikely to accommodate. The enzymes obtained from these efforts function much more slowly than those found in nature, requiring further optimization with the indicated evolution to reach a common rate among natural enzymes.

Recent advances in deep learning have dramatically changed the landscape of computer-based enzyme design. Enzymes can now be generated in almost the same way that AI models such as ChatGPT and Dall-E generate text and images, without the need to use native protein structures to support active sites.

Our team showed that using the structure and amino acid sequence of the active site to stimulate an AI model called RFDiffusion, it could produce the remaining enzyme structures that fully support it. This is equivalent to urging ChatGpt to write an entire short story based on the prompt that it only includes the line “Sadly, the egg never appeared.”

This AI model was used to generate an enzyme called serine hydrolase, a group of proteins with potential uses in medicine and plastic recycling. After designing the enzymes, they were mixed with the intended molecular target to see if the enzymes could catalyze the failure. Encouraged, many of the designs we tested were able to break down molecules and outperformed enzymes previously designed for the same reaction.

To see how accurate the computational design was, we used a method called X-ray crystallography to determine the shape of these enzymes. We found that many of them almost perfectly matched what we designed digitally.

Our findings demonstrate important advances in enzyme design, highlighting how AI can help scientists begin to tackle complex problems. Machine learning tools help more researchers access enzyme design and use the full potential of enzymes to solve modern problems.

This article will be republished from the conversation under a Creative Commons license. Please read the original article.