A new study suggests a potential molecular strategy for treating fragile X syndrome, an inherited neurodevelopmental disorder that causes autism spectrum disorder and intellectual disability. This work shows that enhancing the function of the GluN2b subunit of the N-methyl-D-aspartate (NMDA) receptor signaling pathway can correct key neural dysfunctions in a mouse model of fragile X syndrome.

The research, titled “Non-ionotropic signaling through the NMDA receptor GluN2B carboxy-terminal domain drives dendritic spine plasticity and reverses fragile X phenotypes,” was published in Cell Reports.

Led by Mark Bear, PhD, at MIT’s Picower Institute for Learning and Memory, the study builds on previous work by this group exploring the role of NMDA receptors in regulating synaptic plasticity. The group studies synaptic plasticity and has a history of exploring the molecular basis of fragile X syndrome, and a related disorder, tuberous sclerosis (Tsc).

“We ended up here by accident at the end of the last century, following up on studies of the basic neurobiology of synaptic plasticity. One of our discoveries suggested the possibility that we could reverse aspects of fragile X syndrome, and we have stuck with it for 25 years,” Bear shared with GEN.

Fragile X is characterized by excessive protein synthesis, which leads to synaptic dysfunction and predisposition to seizures, while Tsc involves reduced protein synthesis. In fact, crossbreeding mouse models of both conditions results in healthy offspring, with the protein expression levels balancing each other.

NMDA receptors play a critical role in synaptic plasticity, where calcium ions flow through the receptor and contribute to long-term depression. More recent work from the group identified a signaling pathway for NMDA that was independent of ion flow.

The team hypothesized that two subunits of NMDA, GluN2A and GluN2B, have separate functions contributing to the two functional pathways, with GluN2A contributing to synaptic function via ion flow, while GluN2B modifies protein synthesis through the non-ionotropic mechanism.

To test these hypotheses, the team used the shrinkage and enlargement of dendritic spines as a physical marker for synaptic plasticity in response to modifications in NMDA function. Knockout experiments showed that loss of either GluN2A or GluN2B disrupted long-term depression, a consequence of ionotropic signaling. Knocking out GluN2B eliminated spine shrinkage, a hallmark of synaptic plasticity through non-ionotropic signaling.

Bear commented that by using this method, the team “discovered a novel approach to rebalance altered protein synthesis regulation in these diseases, one that had not been known before.”

The MIT researchers genetically engineered mice in which the carboxy-terminal domain (CTD) of GluN2B was swapped with that of GluN2A. They found that disrupting GluN2B’s CTD eliminated its ability to regulate spine size and increased bulk protein synthesis, mirroring what is seen in fragile X syndrome. Conversely, enhancing GluN2B signaling reduced protein synthesis to normal levels, similar to what is observed in Tsc models.

They then tested whether increasing GluN2B signaling could counteract fragile X-like phenotypes. Fragile X model mice were treated with Glyx-13, which selectively binds to GluN2B. Treatment normalized protein synthesis and reduced seizure susceptibility in fragile X model mice, suggesting that targeting GluN2B could represent a novel therapeutic approach for fragile X syndrome.

All biological function is dependent on how different proteins interact with each other. Protein-protein interactions facilitate everything from transcribing DNA and controlling cell division to higher-level functions in complex organisms.

Much remains unclear, however, about how these functions are orchestrated on the molecular level, and how proteins interact with each other—either with other proteins or with copies of themselves.

Recent findings have revealed that small protein fragments have a lot of functional potential. Even though they are incomplete pieces, short stretches of amino acids can still bind to interfaces of a target protein, recapitulating native interactions. Through this process, they can alter that protein’s function or disrupt its interactions with other proteins.

Protein fragments could therefore empower both basic research on protein interactions and cellular processes, and could potentially have therapeutic applications.

Recently published in Proceedings of the National Academy of Sciences, a new method developed in the Massachusetts Institute of Technology Department of Biology builds on existing artificial intelligence models to computationally predict protein fragments that can bind to and inhibit full-length proteins in E. coli. Theoretically, this tool could lead to genetically encodable inhibitors against any protein.

The work was done in the lab of associate professor of biology and Howard Hughes Medical Institute investigator Gene-Wei Li in collaboration with the lab of Jay A. Stein (1968) Professor of Biology, professor of biological engineering, and department head Amy Keating.

Leveraging machine learning

The program, called FragFold, leverages AlphaFold, an AI model that has led to phenomenal advancements in biology in recent years due to its ability to predict protein folding and protein interactions.

The goal of the project was to predict fragment inhibitors, which is a novel application of AlphaFold. The researchers on this project confirmed experimentally that more than half of FragFold’s predictions for binding or inhibition were accurate, even when researchers had no previous structural data on the mechanisms of those interactions.

“Our results suggest that this is a generalizable approach to find binding modes that are likely to inhibit protein function, including for novel protein targets, and you can use these predictions as a starting point for further experiments,” says co-first and corresponding author Andrew Savinov, a postdoc in the Li Lab. “We can really apply this to proteins without known functions, without known interactions, without even known structures, and we can put some credence in these models we’re developing.”

One example is FtsZ, a protein that is key for cell division. It is well-studied but contains a region that is intrinsically disordered, and therefore, especially challenging to study. Disordered proteins are dynamic, and their functional interactions are very likely fleeting—occurring so briefly that current structural biology tools can’t capture a single structure or interaction.

The researchers leveraged FragFold to explore the activity of fragments of FtsZ, including fragments of the intrinsically disordered region, to identify several new binding interactions with various proteins. This leap in understanding confirms and expands upon previous experiments measuring FtsZ’s biological activity.

This progress is significant in part because it was made without solving the disordered region’s structure, and because it exhibits the potential power of FragFold.

“This is one example of how AlphaFold is fundamentally changing how we can study molecular and cell biology,” Keating says. “Creative applications of AI methods, such as our work on FragFold, open up unexpected capabilities and new research directions.”

Inhibition, and beyond

The researchers accomplished these predictions by computationally fragmenting each protein and then modeling how those fragments would bind to interaction partners they thought were relevant.

They compared the maps of predicted binding across the entire sequence to the effects of those same fragments in living cells, determined using high-throughput experimental measurements in which millions of cells each produce one type of protein fragment.

AlphaFold uses co-evolutionary information to predict folding, and typically evaluates the evolutionary history of proteins using something called multiple sequence alignments (MSAs) for every single prediction run. The MSAs are critical, but are a bottleneck for large-scale predictions—they can take a prohibitive amount of time and computational power.

For FragFold, the researchers instead pre-calculated the MSA for a full-length protein once, and used that result to guide the predictions for each fragment of that full-length protein.

Savinov, together with Keating Lab alumnus Sebastian Swanson, Ph.D., predicted inhibitory fragments of a diverse set of proteins in addition to FtsZ. Among the interactions they explored was a complex between lipopolysaccharide transport proteins LptF and LptG. A protein fragment of LptG inhibited this interaction, presumably disrupting the delivery of lipopolysaccharide, which is a crucial component of the E. coli outer cell membrane essential for cellular fitness.

“The big surprise was that we can predict binding with such high accuracy and, in fact, often predict binding that corresponds to inhibition,” Savinov says. “For every protein we’ve looked at, we’ve been able to find inhibitors.”

The researchers initially focused on protein fragments as inhibitors because whether a fragment could block an essential function in cells is a relatively simple outcome to measure systematically. Looking forward, Savinov is also interested in exploring fragment function outside inhibition, such as fragments that can stabilize the protein they bind to, enhance or alter its function, or trigger protein degradation.

Design, in principle

This research is a starting point for developing a systemic understanding of cellular design principles, and what elements deep-learning models may be drawing on to make accurate predictions.

“There’s a broader, further-reaching goal that we’re building towards,” Savinov says. “Now that we can predict them, can we use the data we have from predictions and experiments to pull out the salient features to figure out what AlphaFold has actually learned about what makes a good inhibitor?”

Savinov and collaborators also delved further into how protein fragments bind, exploring other protein interactions and mutating specific residues to see how those interactions change how the fragment interacts with its target.

Experimentally examining the behavior of thousands of mutated fragments within cells, an approach known as deep mutational scanning, revealed key amino acids that are responsible for inhibition. In some cases, the mutated fragments were even more potent inhibitors than their natural, full-length sequences.

“Unlike previous methods, we are not limited to identifying fragments in experimental structural data,” says Swanson. “The core strength of this work is the interplay between high-throughput experimental inhibition data and the predicted structural models: the experimental data guide us towards the fragments that are particularly interesting, while the structural models predicted by FragFold provide a specific, testable hypothesis for how the fragments function on a molecular level.”

Savinov is excited about the future of this approach and its myriad applications.

“By creating compact, genetically encodable binders, FragFold opens a wide range of possibilities to manipulate protein function,” Li agrees. “We can imagine delivering functionalized fragments that can modify native proteins, change their subcellular localization, and even reprogram them to create new tools for studying cell biology and treating diseases.”

One of nature’s most disliked creatures may very well unlock a breakthrough in disaster response. A multidisciplinary Purdue University research team is recreating mosquito antennae to better study their sensitivity to vibrations. Should the research prove fruitful, it could lead to improvements in monitoring and detecting natural disasters such as earthquakes and tsunamis.

Research groups under Purdue professors Pablo Zavattieri and Ximena Bernal conducted this work, which is published in the journal Acta Biomaterialia.

“We’re still in the early stages but we’re pretty optimistic that we’ll at least learn a great deal,” said Zavattieri, the Jerry M. and Lynda T. Engelhardt Professor of Civil Engineering in Purdue’s College of Engineering. “Taking inspiration from nature and using it to advance scientific research has been a core feature of engineering since the very beginning.”

Despite lacking traditional ears, mosquitoes rely on their antennae to navigate the auditory landscape, homing in on crucial sounds amid the background noise of their own wingbeats.

Through analysis of the mosquitoes’ antennal features—particularly the arrangement and morphology of sensory hairs—civil and construction engineering Ph.D. candidate and team researcher Phani Saketh Dasika said they have already gained profound insights into how these adaptations enhance the auditory sensitivity and selective response to environmental cues.

“Today, we can for all practical purposes read, write, and edit any sequence of DNA, but we cannot compose it. Maybe we can cut and paste pieces from nature’s compositions, but we don’t know how to write the bars for a single enzymatic passage. However, evolution does.” —Frances Arnold, PhD (Nobel Prize Lecture 2018)

Evo, the genome foundation model developed by the Arc Institute published last November that generalizes across the languages of biology — DNA, RNA, and proteins for both predictive and generative capabilities — has received a major update.

In a new preprint that is not yet peer-reviewed and first published on Arc’s website, Evo 2 moves beyond single-cell genomes of bacteria and archaea to include information from humans, plants, and other more complex single-celled and multi-cellular species in the eukaryotic domain of life.

The model’s resulting research applications span a diverse array of scientific fields including drug discovery, agriculture, industrial biotechnology, and material science. The multimodal and multiscale work is a collaboration with Nvidia along with contributors from Stanford University, UC Berkeley, and UC San Francisco.

“The recipe for life is entirely present in the genetic information contained in our DNA,” said Kimberly Powell, vice president of healthcare at Nvidia. “We’re seeking a deeper understanding of biological complexity. Evolution has solved this problem over millions of years, and Evo 2 aims to learn from this knowledge.”

In healthcare, understanding which gene variants are tied to a disease is an invaluable tool for therapeutics. Early validation of Evo 2’s capabilities showed that the model can identify how genetic mutations affect protein, RNA, and organismal fitness. In tests with variants of BRCA1, a gene associated with breast and ovarian cancer risk, Evo 2 achieved greater than 90% accuracy in predicting which mutations are benign versus disease-causing. 

Patrick Hsu, PhD, Arc Institute co-founder and an assistant professor of bioengineering at UC Berkeley, stated that Evo 2 is the only model that can predict the effects of both coding and noncoding mutations.

“It is the second-best model for coding mutations, but it is state-of-the-art for noncoding mutations, which other variant effect prediction methods, such as AlphaMissense from DeepMind, cannot score,” said Hsu. 

Hsu also described Evo 1 as a “blurry picture of single-cell life” because it was trained on a corpus of 300 billion nucleotides derived from prokaryotic genomes. The team “wanted to be much more ambitious” in this collaboration with Nvidia. 

Evo 2 was built on NVIDIA’s DGX Cloud platform and is trained on more than 9.3 trillion nucleotides from the genomes of more than 128,000 species across the tree of life. The model uses a novel architecture called StripedHyena 2, which enabled training that was “nearly three times faster than optimized transformer models,” according to Dave Burke, PhD, chief technology officer at Arc Institute. The model also has 40 billion parameters and is similar in scale to the current generation of large language models released from Meta, DeepMind, or OpenAI. 

Evo 2 can process DNA sequences of up to 1 million nucleotides at once, allowing it to understand relationships between distant parts of the genome. Hsu stated that this long context length unlocks multiple molecular scales, from short biological molecules, such as tRNA, or clusters of genes (e.g., operons), to entire bacterial genomes or eukaryotic chromosomes. 

In principle, mRNA vaccines are ideal for health emergencies as they can be quickly mass produced using a template. The problem is that current cell-based template production methods take too long, says pandemic preparedness organization, CEPI.

And slow template production limits access to vaccines, according to Chaminda Salgado, CMC technology lead, who told GEN about CEPI’s efforts to develop an alternative with French technology firm, DNA script.

“While the process of producing mRNA itself takes only around seven days, the creation of a DNA template can take up to a month, creating a bottleneck in the manufacturing process. That’s because creating a DNA template typically involves living organisms, such as bacteria or yeast, to grow, extract, and purify the required DNA plasmid templates.

“This new project aims to overcome these challenges by generating automated synthetic DNA templates, which don’t need to be grown in living cells and can be rapidly produced within days rather than weeks, without the need for expensive biopharmaceutical facilities or highly trained staff to run the process,” Salgado says.

The plan is to take DNA oligonucleotides made using DNA Script’s enzyme driven synthesis technology—called Syntax—and combine them into longer template sequences using an automated “gene assembler” system.

Manufacturing cost

Replacing cell culture-based template production with synthetic methods could have a significant impact on manufacturing cost and timelines, according to Salgado.

“By streamlining the process, mRNA vaccines could be manufactured and developed even faster and cheaper, meaning vaccines could be given to at-risk populations sooner and potentially help to stop an outbreak in its tracks.

“This technology supports CEPI’s 100 Days Mission—a goal to produce pandemic-busting vaccines within 100 days of a viral threat being identified—and complements a suite of CEPI investments aiming to increase the speed at which vaccines can be manufactured,” he adds.

Another aim of the partnership is to make it easier for people to access vaccines by enabling production in areas without an established manufacturing infrastructure.

Salgado says, “This partnership supports equitable access as it reduces the need for expensive pharmaceutical equipment and infrastructure associated with traditional biologically manufactured DNA templates.

“If successful, the lower costs associated with automated, synthetic DNA template production would remove a potential barrier to the introduction of the technology in Global South countries where resources may be more limited. This could enable the possibility of vaccination closer to the site of an outbreak and avoid delays in vaccine access.”

CEPI and DNA Script are also committed to enabling equitable access to the outputs of the collaboration, Salgado says, adding, “This ultimately includes a commitment to vaccines being available first to populations at risk when and where they are needed at an affordable price should a related vaccine be developed further using CEPI funding.

“Project results, including data generated as part of this project, will be published open access for the benefit of the global scientific community.”

University of Queensland researchers have for the first time introduced genetic material into plants via their roots, opening a potential pathway for rapid crop improvement.

Professor Bernard Carroll from UQ’s School of Chemistry and Molecular Biosciences said nanoparticle technology could help fine-tune plant genes to increase crop yield and improve food quality.

“Traditional plant breeding and genetic modification take many generations to produce a new crop variety, which is time-consuming and expensive,” Professor Carroll said.

“We have succeeded in having plant roots absorb a benign nanoparticle which was developed by Professor Gordon Xu’s group at UQ for the delivery of vaccines and cancer treatments in animals.

“Plant cell walls are rigid and wood-like, much tougher than human or animal cells so we coated the nanoparticle with a protein that gently loosens the plant cell wall.

“The protein coating helped the nanoparticle break through the cell walls to deliver a synthetic mRNA cargo into plants for the first time.”

mRNAs are natural messenger molecules containing genetic instructions to build and enhance all forms of life.

The research team used the nanoparticles to deliver synthetic mRNA that produces a green fluorescent protein into multiple plant species including Arabidopsis, a miniature member of the canola and cabbage family used extensively in genetic research.

“It was surprising that rather than delivering all of its load in the first cell it entered, the nanoparticle travelled with water through the plant distributing the mRNA as it went,” Professor Carroll said.

“This is exciting because with further improvement, the technology could potentially be used in the future to produce new crop varieties more quickly.

“With further research we could target an issue with a crop such as flavour or quality and have a new variety without the need for a decade of cross breeding or genetic modification.

“Similar to how an mRNA vaccine produces a protein to stimulate the immune system and then degrades away, the mRNA we deliver into plants is expressed transiently and then disappears.”

The nanoparticle technique has been patented by UQ’s commercialisation company UniQuest, which is now seeking partners to further develop the technology.

The research team included Professor Zhi Pin (Gordon) Xu and Dr Jiaxi Yong at UQ’s Australian Institute for Bioengineering and Nanotechnology and Queensland Alliance for Agriculture and Food Innovation.

Scientists have developed a breakthrough method to track diet using stool metagenomic data.

Developed by researchers at the Institute for Systems Biology (ISB), the new method, called MEDI (Metagenomic Estimation of Dietary Intake), detects food-derived DNA in stool samples to estimate dietary intake.

MEDI leverages stool metagenomics, which refers to sequencing all the DNA present in fecal samples (including microbial, human, and food-derived DNA). This non-invasive, data-driven approach offers an objective alternative to traditional food diaries and questionnaires, which are still the gold standard in dietary assessment but can suffer from misreporting and compliance issues.

“For decades, nutrition research has depended on self-reported diaries and questionnaires — approaches that require a high degree of effort and compliance from research participants. How many strawberries did I eat two days ago? Did I have one glass of orange juice with breakfast, or two?” said Dr. Christian Diener, lead author of the study.

“MEDI provides a solution by analyzing food-derived DNA in gut metagenomic samples, offering a convenient alternative that shows good agreement with known dietary and nutritional intake patterns.”

Key Findings:

• An Alternative to Questionnaire-Based Diet Tracking: Leveraging a database of more than 400 food items and over 300 billion base pairs of genomic information, MEDI accurately detected food intake patterns in infants and adults, and across two controlled feeding studies. MEDI
• Connecting Dietary Intake to Nutrition: MEDI converts the relative abundance profile of specific food items into nutrient profiles, assuming a 100 gram portion. These nutrient profiles show good agreement with data from controlled feeding studies.
• Identified Diet-Related Health Risks: Without food logs, MEDI pinpointed dietary features linked to metabolic syndrome in a large clinical cohort.

“Our study represents a major leap forward in how we track diet and its impacts on human health,” said ISB Associate Professor Dr. Sean Gibbons, senior author of the study.

“With food-derived DNA signatures in stool, we now have a powerful way to measure diet and microbiome composition from the same sample, which will expand our understanding of the forces shaping the human gut microbiome, personalized nutritional responses, and disease risk.”

With further development, MEDI could transform nutrition science, epidemiological studies, and clinical trials, allowing researchers, doctors, and individuals to track diet-related health risks with unprecedented ease.

The CRISPR-Cas gene scissors offer a wide range of potential applications, from the treatment of genetic diseases to antiviral therapies and diagnostics. However, to safely harness their powers, scientists are searching for mechanisms that can regulate or inhibit the systems’ activity. Enter the anti-CRISPR protein AcrVIB1, a promising inhibitor whose exact function has remained a mystery—until now.

A research team from the Helmholtz Institute for RNA-based Infection Research (HIRI) in Würzburg, in collaboration with the Helmholtz Centre for Infection Research (HZI) in Braunschweig, has uncovered the precise way AcrVIB1 works that expands the known means by which Acrs can shut down CRISPR. The results are published in the journal Molecular Cell.

Bacteria and their viruses, known as phages, are locked in an age-old arms race. To defend against phage attacks, bacteria have evolved sophisticated mechanisms to recognize and counteract invading viruses. In turn, phages have developed innovative strategies to evade these defenses. A prime example of this ongoing battle is the CRISPR-Cas defense system in bacteria, countered by anti-CRISPR proteins (Acrs) in phages, which specifically block these bacterial “gene scissors.”

Apart from their counter-defensive function, anti-CRISPR proteins hold great promise for enabling more precise control over CRISPR technologies. The research team has now further elucidated the function of an important yet so far uncharacterized anti-CRISPR protein.

“In a previous study, we used a deep learning algorithm to predict new Acrs. This led to the identification of AcrVIB1, the first anti-CRISPR protein targeting the Cas13b nuclease,” says HIRI department head Prof. Chase Beisel, who led the study together with the department of Prof. Wulf Blankenfeldt at HZI.

“The nuclease Cas13b can recognize and cut RNA. It is currently used to silence genes, whether to study their function, clear viruses, or counteract genetic diseases linked to the gene.” However, how the protein AcrVIB1 inhibits Cas13b remained unknown until now. In their study, the research team presents this entirely new blocking mechanism.

An RNA dead end

The Cas13b nuclease operates by interacting with a CRISPR ribonucleic acid (crRNA), which serves as a guide to identify and bind to complementary RNA sequences, for example those from phages. Once the target RNA is bound, Cas13b can cleave and degrade not only these complementary RNA molecules but also all other RNAs in the vicinity.

While most known anti-CRISPR proteins block steps along this path such as crRNA binding or target recognition, AcrVIB1 adopts a radically different strategy: Rather than blocking the binding of the crRNA to Cas13b, AcrVIB1 even improves it. The formed pair is dysfunctional though, meaning that the enzyme cannot begin degrading RNAs even when its target is present. Furthermore, the bound crRNA becomes vulnerable to attack by cellular ribonucleases, which break down RNA molecules.

“The tighter binding between nuclease and guide RNA was entirely unexpected. The simpler and therefore initially expected mechanism would have been to just prevent the guide RNA from binding in the first place,” says first author Dr. Katharina Wandera, who completed her doctorate in Chase Beisel’s laboratory.

“Nevertheless, the path taken by AcrVIB1 appears to be more effective: AcrVIB1 binds tightly to and thereby renders Cas13b inactive. At the same time, it increases the turnover of guide RNAs, making Cas13b a dead end for crRNAs.”

Chase Beisel’s team at HIRI and the laboratory of Wulf Blankenfeldt at HZI have joined forces to decipher the structure of the inhibition mechanism more precisely. Using cryo-electron microscopy, Blankenfeldt’s group showed that AcrVIB1 binds to Cas13b, leaving the crRNA-binding domain untethered.

“Our finding provides a blueprint for the development of molecules that could mimic or modify the function of the anti-CRISPR protein,” says Blankenfeldt. These are the first data from the HZI’s new cryo-electron microscopy facility to be published.

Enzymes are molecular machines that carry out the chemical reactions that sustain all life, an ability that has captured the attention of scientists like me.

Consider muscle movement. Your body releases a molecule called acetylcholine to trigger your muscle cells to contract. If acetylcholine sticks around for too long, it can paralyze your muscles—including your heart muscle cells—and, well, that’s that. This is where the enzyme acetylcholinesterase comes in. This enzyme can break down thousands of acetylcholine molecules per second to ensure muscle contraction is stopped, paralysis avoided and life continued. Without this enzyme, it would take a month for a molecule of acetylcholine to break down on its own—about 10 billion times slower.

You can imagine why enzymes are of particular interest to scientists looking to solve modern problems. What if there were a way to break down plastic, capture carbon dioxide or destroy cancer cells as fast as acetylcholinesterase breaks down acetylcholine? If the world needs to take action quickly, enzymes are a compelling candidate for the job—if only researchers could design them to handle those challenges on demand.

Designing enzymes, unfortunately, is very hard. It’s like working with an atom-sized Lego set, but the instructions were lost and the thing won’t hold together unless it’s assembled perfectly. Newly published research from our team suggests that machine learning can act as the architect on this Lego set, helping scientists build these complex molecular structures accurately.

What’s an enzyme?

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

Enzymes are proteins—large molecules that do the behind-the-scenes work that keep all living things alive. These proteins are made up of amino acids, a set of building blocks that can be stitched together to form long strings that get knotted up into specific shapes.

The specific structure of a protein is key to its function in the same way that the shapes of everyday objects are. For example, much like a spoon is designed to hold liquid in a way that a knife simply can’t, the enzymes involved in moving your muscles aren’t well suited for photosynthesis in plants.

For an enzyme to function, it adopts a shape that perfectly matches the molecule it processes, much like a lock matches a key. The unique grooves in the enzyme—the lock—that interact with the target molecule—the key—are found in a region of the enzyme known as the active site.

The active site of the enzyme precisely orients amino acids to interact with the target molecule when it enters. This makes it easier for the molecule to undergo a chemical reaction to turn into a different one, making the process go faster. After the chemical reaction is done, the new molecule is released and the enzyme is ready to process another.

How do you design an enzyme?

Scientists have spent decades trying to design their own enzymes to make new molecules, materials or therapeutics. But making enzymes that look like and go as fast as those found in nature is incredibly difficult.

Enzymes have complex, irregular shapes that are made up of hundreds of amino acids. Each of these building blocks needs to be placed perfectly or else the enzyme will slow down or completely shut off. The difference between a speed racer and slowpoke enzyme can be a distance of less than the width of a single atom.

Initially, scientists focused on modifying the amino acid sequences of existing enzymes to improve their speed or stability. Early successes with this approach primarily improved the stability of enzymes, enabling them to catalyze chemical reactions at a higher range of temperatures. But this approach was less useful for improving the speed of enzymes. To this day, designing new enzymes by modifying individual amino acids is generally not an effective way to improve natural enzymes.

Researchers found that using a process called directed evolution, in which the amino acid sequence of an enzyme is randomly changed until it can perform a desired function, proved much more fruitful. For example, studies have shown that directed evolution can improve chemical reaction speed, thermostability, and even generate enzymes with properties that aren’t seen in nature. However, this approach is typically labor-intensive: You have to screen many mutants to find one that does what you want. In some cases, if there’s no good enzyme to start from, this method can fail to work at all.

Both of these approaches are limited by their reliance on natural enzymes. That is, restricting your design to the shapes of natural proteins likely limits the kinds of chemistry that enzymes can facilitate. Remember, you can’t eat soup with a knife.

Is it possible to make enzymes from scratch, rather than modify nature’s recipe? Yes, with computers.

Designing enzymes with computers

The first attempts to computationally design enzymes still largely relied on natural enzymes as a starting point, focusing on placing enzyme active sites into natural proteins.

This approach is akin to trying to find a suit at a thrift store: It is unlikely you will find a perfect fit because the geometry of an enzyme’s active site (your body in this analogy) is highly specific, so a random protein with a rigidly fixed structure (a suit with random measurements) is unlikely to perfectly accommodate it. The resulting enzymes from these efforts performed much more slowly than those found in nature, requiring further optimization with directed evolution to reach speeds common among natural enzymes.

Recent advances in deep learning have dramatically changed the landscape of designing enzymes with computers. Enzymes can now be generated in much the same way that AI models such as ChatGPT and DALL-E generate text or images, and you don’t need to use native protein structures to support your active site.

Our team showed that when we prompt an AI model, called RFdiffusion, with the structure and amino acid sequence of an active site, it can generate the rest of the enzyme structure that would perfectly support it. This is equivalent to prompting ChatGPT to write an entire short story based on a prompt that only says to include the line “And sadly, the eggs never showed up.”

We used this AI model specifically to generate enzymes called serine hydrolases, a group of proteins that have potential applications in medicine and plastic recycling. After designing the enzymes, we mixed them with their intended molecular target to see whether they could catalyze its breakdown. Encouragingly, many of the designs we tested were able to break down the molecule, and better than previously designed enzymes for the same reaction.

To see how accurate our computational designs were, we used a method called X-ray crystallography to determine the shapes of these enzymes. We found that many of them were a nearly perfect match to what we digitally designed.

Our findings mark a key advance in enzyme design, highlighting how AI can help scientists start to tackle complex problems. Machine learning tools could help more researchers access enzyme design and tap into the full potential of enzymes to solve modern-day problems.