The study was led by Professor Rachel Kolodny and PhD candidate Guy Yanai of the University of Haifa, together with Professor Nir Ben-Tal and graduate student Gabriel Axel of Tel Aviv University, and Specially Appointed Associate Professor Liam M. Longo of ELSI. Kolodny also spent five months as a visiting researcher at ELSI, developing methods to analyze the new model. Their creation, dubbed CLSS for Contrastive Learning Sequence-Structure, is a protein language model designed to overcome a stubborn problem that has limited previous computational approaches: the awkward relationship between what a protein’s sequence says and what its structure actually does.

Scientists have long organized proteins into hierarchical groups based on relatedness, much like the genus and species categories biologists use to classify organisms. These curated systems, such as the widely used ECOD and CATH databases, distill decades of expert knowledge. But with artificial intelligence now capable of generating ’embeddings’—numerical representations in which proteins with similar properties receive nearby coordinates, like postal codes on a map—researchers can visualize relationships across millions of proteins at once, producing what the team calls a protein world map. The catch is that sequence and structure do not map neatly onto each other. Unrelated sequences can fold into similar shapes, while even identical sequences can sometimes adopt wildly different structures.

Most existing protein language models treat sequence and structure as separate worlds, processing one or the other independently. Even hybrid models that incorporate both kinds of data rarely place the sequence and the structure of the same protein at the same location on a global map, leaving researchers with two conflicting atlases of protein space. CLSS was engineered specifically to resolve this discordance. Using a machine learning strategy known as contrastive learning, the model is trained on pairs of protein sequences and their corresponding structures, learning to pull matching sequence-structure pairs together in the embedding space while pushing unrelated pairs apart.

The result is a single shared map in which a protein occupies essentially the same location whether the model is given its sequence or its structure. When benchmarked against other state-of-the-art protein language models, CLSS succeeded in producing a cohesive unified representation, something its predecessors could not achieve. Remarkably, the model’s maps closely reproduced the relationships recorded in the expert-curated ECOD and CATH classification systems, even though those classifications were never shown to the model during training. In direct classification tests, CLSS also performed strongly, demonstrating that merging sequence and structure information yields genuinely more informative protein representations.

Perhaps the most exciting feature of CLSS is its ability to handle fragments. Most protein language models require a complete sequence or structure to generate a meaningful embedding, but CLSS showed that short sequence fragments can in many cases be positioned meaningfully alongside full-length proteins and structures. This capability matters enormously for evolutionary studies, because small pieces of proteins have been repeatedly reused and rearranged throughout the history of life. Some fragments may even have served as the primordial building blocks from which the earliest protein domains were assembled, meaning that similar fragments appearing in otherwise unrelated proteins can hint at ancient evolutionary connections.

The maps produced by CLSS also revealed sweeping patterns across protein space that were previously difficult to see. When the researchers overlaid biological properties onto the maps, proteins associated with organic cofactors turned out to cluster in particular regions, while metal-binding proteins were scattered more broadly. Such patterns illustrate how global protein maps can serve not only as classification tools but as instruments for exploring the interplay between sequence, structure, function, and deep evolutionary history, potentially exposing large-scale patterns invisible to conventional pairwise comparison methods.

‘This gives us a way to look at the protein universe through sequence and structure at the same time, rather than treating them as separate worlds,’ said Longo. ‘What is particularly exciting for us is the possibility of using these maps to uncover large-scale evolutionary patterns that are difficult to recognise using conventional approaches.’ The team ultimately envisions unified sequence-structure representations opening new frontiers in database searches, protein engineering, and the reconstruction of evolutionary trajectories—offering a fresh window onto how the staggering diversity of proteins found in life today emerged over nearly four billion years of evolution.

Subject of Research: A contrastive-learning protein language model that unifies protein sequence and structure representations to map the protein universe

Article Title: Uniting sequence and structure to map the protein universe

Article References: Uniting sequence and structure to map the protein universe. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: protein language model, CLSS, protein evolution, contrastive learning, protein structure, amino acid sequence, ECOD, CATH, protein embeddings, evolutionary biochemistry, protein classification, Institute of Science Tokyo

Gasdermins occupy a unique position in the architecture of innate immunity. The family, which in humans includes GSDMA, GSDMB, GSDMC, GSDMD and GSDME, shares a common operational logic. Each protein consists of a cytotoxic N-terminal domain tethered to a C-terminal domain that acts as an internal restraint. In the resting state, the two domains bind each other so that the pore-forming capacity of the N-terminus is masked. When pattern-recognition receptors detect pathogen-associated or damage-associated molecular patterns, they trigger proteolytic cascades—inflammasome assemblies that activate inflammatory caspases such as caspase-1, caspase-4, caspase-5 and caspase-11, and, in apoptotic contexts, caspase-3 or granzyme-mediated cleavage. These enzymes cut the gasdermin at a flexible linker region, releasing the N-terminal fragment. Freed from its autoinhibitory partner, the fragment translocates to the plasma membrane, oligomerizes and inserts a large beta-barrel pore with an inner diameter on the order of 10 to 20 nanometers.

The consequences of pore formation are dramatic and rapid. Ions rush down their electrochemical gradients, water follows osmotically, the cell swells and bursts in the lytic mode of death known as pyroptosis. Before rupture, the pores permit the efflux of potassium and the release of mature interleukin-1beta and interleukin-18, alarmins such as high-mobility group box 1, and other inflammatory cargo that summon and shape the immune response. Pyroptosis is therefore not merely a demolition but a broadcast: the dying cell converts its own destruction into an alarm signal that recruits neutrophils, activates antigen-presenting cells and biases the adaptive immune system toward antiviral effector programs. This dual character—killing and alerting—makes the gasdermin system attractive for therapeutic exploitation, particularly against pathogens that thrive by suppressing or evading conventional immune pathways.

Viruses and gasdermins have long been adversaries in an evolutionary arms race. Many viruses encode inhibitors that block inflammasome sensors, sequester gasdermin fragments or interfere with caspase activation, reflecting the selective pressure that pyroptosis exerts on viral replication. Poxviruses, herpesviruses, influenza viruses and coronaviruses all deploy strategies to dampen inflammatory cell death. Conversely, host cells can route viral sensing signals toward gasdermin activation through multiple sensors, including ZBP1, which detects influenza A virus through recognition of Z-form nucleic acid, and AIM2 or IFI16, which sense foreign DNA. The observation that gasdermin activation can restrict viral replication even when interferon responses are disabled underscores the pathway’s value as a fail-safe. The Cell Research article situates the new engineering efforts within this biological context, arguing that the natural system’s potency has been limited chiefly by its lack of specificity and by viral countermeasures.

Here artificial intelligence changes the calculus. Modern structure-prediction systems such as AlphaFold2 and its successors have resolved the atomic architectures of gasdermin domains, their autoinhibited complexes and their membrane-inserted oligomeric pores, giving designers an accurate map of the conformational switch that governs activity. More consequentially, diffusion-based and language-model-based protein design tools now allow researchers to specify a desired function—a binding interface, a cleavage site, a regulatory logic module—and generate amino acid sequences predicted to fold into structures that fulfill it. Rather than screening natural variants or making incremental mutations, designers can compose gasdermin-based molecules from the ground up, fusing pore-forming domains to sensor modules that respond to molecular features found only in infected cells.

The design logic described in the article follows a gating principle. An engineered construct remains inert until it encounters a virus-specific cue: a viral protease that cleaves a designed linker, a viral RNA or DNA species bound by an engineered sensor domain, or a host-state marker such as a receptor induced by interferon signaling. Only when the gate opens is the gasdermin N-terminal domain released or reconfigured to oligomerize at the membrane. In principle, such constructs could discriminate infected from uninfected tissue with a fidelity that natural inflammasome pathways, which respond to broad danger signals, cannot achieve. The article emphasizes that computational modeling of pore geometry, oligomerization energetics and membrane interactions is essential at every step, because even small deviations in the N-terminal domain can abolish pore formation or, conversely, produce toxic nonspecific membrane binding.

Experimental validation of AI-designed gasdermins, as discussed in the piece, proceeds through iterative cycles in which predicted structures are tested in liposome leakage assays, cell-culture infection models and, ultimately, animal studies. Key metrics include the tightness of the off state, the sensitivity and specificity of the trigger response, the efficiency of membrane pore formation and the immunological consequences of pyroptotic lysis in vivo. The authors highlight that design failures are informative: constructs that leak activity reveal the energetic margins of autoinhibition, while constructs that fail to activate expose weaknesses in sensor-linker coupling. Each cycle feeds data back into the design pipeline, a workflow that has already accelerated progress in other classes of engineered proteins, including designed cytokines, antibody mimetics and switchable cell-death regulators.

The therapeutic implications extend across antiviral medicine and beyond. A programmable gasdermin could, in principle, be delivered as a gene therapy or mRNA therapeutic to tissues vulnerable to a specific pathogen, standing ready to eliminate infected cells before viral spread becomes established. Such an approach would be particularly valuable against emerging viruses for which vaccines and antivirals lag behind outbreak speed, and against chronic infections where viral evasion of immune clearance is the central obstacle. The same design principles could be adapted to oncology, since many tumors evade pyroptosis by silencing gasdermin expression or downregulating inflammasome components, and engineered constructs triggered by tumor-specific proteases or neoantigens could restore an inflammatory form of cancer-cell death that promotes antigen release and immune priming. The article notes that the concept of AI-guided cell-death engineering generalizes: gasdermins are one member of a broader class of pore-forming effectors, including the immune proteins MLKL in necroptosis and the complement membrane-attack complex, whose activity might similarly be placed under synthetic control.

Substantial challenges temper the enthusiasm. Uncontrolled pyroptosis is dangerous: excessive gasdermin activation drives cytokine storms, tissue damage and septic shock, as demonstrated by the lethal inflammation observed when gasdermin pores open unchecked during severe infections. Any engineered system must therefore incorporate fail-safes, such as dependence on multiple simultaneous triggers, dose-limiting delivery strategies and pharmacological off switches. Immune responses against designed protein sequences pose another obstacle, as does the difficulty of achieving tissue-restricted expression. Off-target activation in bystander cells, even at low frequency, could produce disproportionate inflammation given the potency of the pore-forming mechanism. The authors stress that computational predictions, however accurate, must be paired with rigorous empirical safety testing across diverse cell types and inflammatory contexts before clinical translation can be contemplated.

Nevertheless, the trajectory is clear. The gasdermin system, once understood only as a blunt instrument of innate defense, is becoming a designable platform. Artificial intelligence supplies the structural insight and generative capacity to reprogram its trigger logic, its target selectivity and even its pore properties, while the underlying biology supplies a death mechanism that is fast, inflammatory and inherently immunogenic—qualities well suited to antiviral defense. The Cell Research article frames this convergence as the beginning of a programmable immunology, in which the executioners of cell death are no longer merely triggered by infection but are deliberately built to seek it out. If the engineering challenges of specificity, control and delivery can be met, AI-guided gasdermins may transform the treatment of viral disease from a defensive struggle into a precise, pre-emptive campaign against the cells that harbor the enemy.

Subject of Research: AI-guided engineering of gasdermin proteins to induce pyroptotic death of virus-infected cells

Article Title: Programmed to kill: AI-guided gasdermins eliminate virus-infected cells

Article References: Betrancourt, A., & Broz, P. (2026). Programmed to kill: AI-guided gasdermins eliminate virus-infected cells. Cell Research. https://doi.org/10.1038/s41422-026-01292-y

Image Credits: AI Generated

DOI: 10.1038/s41422-026-01292-y

Keywords: gasdermin, pyroptosis, AI protein design, innate immunity, virus-infected cells, inflammasome, cell death, antiviral therapy, protein engineering, pore-forming proteins, Cell Research, immunotherapy

The team, led by senior author Daniel G. Isom, Ph.D., a Sylvester researcher and faculty member in the Department of Molecular and Cellular Pharmacology, turned to a vast computational dataset containing more than 214 million predicted protein structures. Rather than scanning for familiar sequence signatures, the researchers asked a different question: which of these predicted proteins fold into shapes that resemble known functional families, even if their sequences look like nothing recognizable? Their search concentrated on the G protein-coupled receptor family, or GPCRs, an enormous group of membrane proteins that allow cells to sense and respond to signals arriving from outside the cell. GPCRs are among the most heavily exploited targets in pharmacology, so any undiscovered relatives are of more than academic interest.

“For decades, we have largely explored protein biology using sequence as our guide,” Isom said. “We wanted to know what biology we might be missing if we searched by three-dimensional structure instead. What we found suggests there is another layer of biology that has been hiding in plain sight.” The structural approach flagged a set of proteins that classical sequence-based methods had never connected to the GPCR family, a category sometimes described as part of the dark proteome, the substantial fraction of predicted proteins whose functions remain entirely unknown.

One protein in particular stood out. Known as TM184C, it folded like a GPCR, but when the team examined its behavior in cells, it refused to follow the expected script. Classic GPCRs typically sit in the plasma membrane at the cell’s surface, waiting for extracellular ligands. TM184C, by contrast, was found largely inside the cell, embedded in the membranes of intracellular vesicles, the tiny membrane-bound packages that cells use to ferry materials between compartments and to one another. That difference in localization alone hinted that TM184C might represent an entirely new mode of GPCR-like function.

Following the protein’s location in living cells revealed something stranger still. The TM184C-positive vesicles did not sit still. They traveled along microtubules, the rigid protein filaments that serve as the cell’s internal highway system, and they accumulated in thin projections that extend from one cell toward its neighbors. These projections acted like bridges. Through them, the researchers observed cells exchanging metabolites, vesicles, and even entire organelles, including mitochondria, the power-generating structures that supply the energy currency of life. The discovery suggests that direct, physically connected exchange between neighboring cells may be far more common, and far more consequential, than previously appreciated.

“When we saw TM184C-positive vesicles moving through connections between cells, we realized these structures could be routes for substantial material exchange,” said Jenniffer Arcuri, Ph.D., a senior scientist with Sylvester and the study’s lead author. “That completely changed how we thought about TM184C and made us consider how cells might use these connections to cooperate and compete for resources.” To test whether the protein actually mattered, the team disrupted TM184C in cultured cells. The effect was clear: the cells formed fewer intercellular connections, and their overall shape and vesicle organization changed, indicating that TM184C helps build and manage these intercellular conduits rather than merely riding along inside them.

The findings raise a question that cuts to the heart of tissue biology: when neighboring cells share resources, who benefits? In healthy tissue, the exchange could be a form of cooperation, allowing cells under stress to survive by shuttling fuel, building blocks or damaged components to wherever they are needed most. But the conduits could also be exploited. If the exchange is unequal, one cell might gain at another’s expense, drawing support from a weaker neighbor. That possibility becomes especially provocative in cancer, where tumor cells frequently endure low oxygen and scarce nutrients. Intercellular bridges could give some cancer cells a lifeline, allowing them to share resources or siphon support from surrounding tissue in ways that conventional metabolic studies, which typically analyze cells in isolation, would never detect.

“I think cells coordinate until they have to compete,” said Shraddha Chandthakuri, a Cancer Biology doctoral student in the Isom lab. “When the cells are stressed, they may coordinate to redistribute the proteins, organelles, and metabolites to support the survival of the population as a whole.” Bruno Colon, a Molecular and Cellular Pharmacology graduate student in the same lab, is now probing that dynamic directly. “What excites me most is understanding what this exchange actually does to the cells on both sides,” Colon said. “As part of my doctoral work in the Isom lab, I am studying how these connections occur in normal cells and aggressive cancers like glioblastoma. Understanding their role could give us new insight into how these tumors communicate and potentially reveal vulnerabilities we haven’t recognized before.”

TM184C appears to influence more than just physical connectivity. The protein also seems to help regulate autophagy, the recycling program by which cells break down and reuse old or damaged components, a process critical to surviving starvation and other stresses. In the study, when the researchers reduced the amount of TM184C in cells, markers of autophagy rose, suggesting the protein normally acts as a brake or tuning mechanism on the process. Adding a structural dimension to the evidence, the team studied a yeast protein called Hfl1 that resembles the human protein. When Hfl1 was removed from yeast, the cells developed noticeable defects. Remarkably, inserting human TM184C into those yeast rescued the problems, demonstrating that the protein’s essential function has been conserved across roughly a billion years of evolution separating baker’s yeast from humans.

For Isom, the broader lesson is about method as much as mechanism. He emphasized that the work depended on pairing AI-driven structure prediction with rigorous experimental validation, not on trusting the algorithms alone. “AI cannot be blindly trusted, but can lead to really big things in the hands of experts and prepared minds,” he said. “For decades, biomedical research has understandably concentrated on the proteins we could identify and understand. But there is another layer of biology that has remained largely invisible to us. AI gives us a way to start exploring it systematically. TM184C is one example of what can be found when we look.” The implication is that artificial intelligence is not merely accelerating the pace of existing science; it is changing what scientists can see at all. TM184C, pulled out of the dark proteome by searching the shapes rather than the sequences of 214 million predicted proteins, offers both a new way to study how cells communicate, survive stress and possibly drive disease, and a template for finding whatever else has been hiding in plain sight.

Subject of Research: Discovery and functional characterization of the hidden GPCR-like human protein TM184C, which regulates intercellular exchange and autophagy

Article Title: AI helps find hidden human proteins and reveals what they do

Article References: AI helps find hidden human proteins and reveals what they do. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: TM184C, GPCR, dark proteome, artificial intelligence, protein structure prediction, intercellular exchange, autophagy, vesicles, mitochondria, cell-to-cell bridges, Sylvester Comprehensive Cancer Center, cancer biology

Addressing this significant gap, computational biologists at Emory University have introduced an innovative approach that sets out to quantify the trustworthiness of protein language model embeddings. Published in Nature Methods, their novel framework evaluates the quality of the model’s internal representations by contrasting embeddings of natural proteins with those generated from synthetic, random sequences. This comparative strategy enables researchers to discern how confidently the model distinguishes biologically meaningful signals from noise, marking a transformative step in understanding and validating AI-driven biological inferences.

Embeddings refer to the numerical representation language models assign to data, condensing complex inputs into abstract vectors within a latent space where proximity implies similarity. In protein language models, this latent space metaphorically catalogues protein sequences based on structural and functional features discerned during training. Emory’s team visualized this latent space as a scatter plot, observing that natural proteins cluster according to their evolutionary and functional subtypes, while synthetic sequences, devoid of biological relevance, occupy distinctly separate regions. They coined this latter region the “junkyard,” positing it as a repository of low-quality embeddings that reflect the model’s unfamiliarity with non-biological sequences.

Central to their methodology is the concept of a “random neighbor score,” a metric that quantifies the proximity between a given protein’s embedding and those of synthetic, random sequences within the latent space. A low score indicates few or no synthetic neighbors nearby, suggesting the model’s high confidence in the biological validity of that protein’s embedding. Conversely, a high score implies the embedding is closer to the “junkyard,” signaling uncertainty or diminished reliability. This quantitative measure provides a computationally efficient and biologically grounded proxy for evaluating the model’s predictive certainty across diverse proteins and subsequences.

The implications of this development reach far beyond theoretical considerations. Protein sequences, encoded by DNA, fold into intricate three-dimensional structures that underpin nearly all cellular processes—from catalysis and signaling to defense mechanisms. With over 200 million protein sequences cataloged in databases such as UniProt, language models have a substantial foundation for training. Still, the true diversity of proteins extends into the trillions, much of it residing in the enigmatic microbial world that community metagenomes represent. Emory’s framework offers a vital means to assess whether inferences drawn from limited sample sets can generalize reliably to this vast, largely uncharted biosphere.

The endeavor to understand metagenomic complexity is vital because microorganisms do not exist in isolation; they form dynamic communities that profoundly impact host health and ecosystem functions. Characterizing the proteins encoded by these communities uncovers biochemical pathways and interactions that conventional experimental methods struggle to elucidate at scale. By sharpening the lens through which AI models interpret protein sequences, the newly introduced confidence metric enhances the promise of computational biology to unlock unprecedented biological insights.

This advancement also illuminates the intricate process of evolution etched into protein sequences. Evolution conserves amino acid residues essential for a protein’s function, imprinting a signature that language models learn to recognize during training. Natural proteins, therefore, exhibit coherent embedding patterns, reflective of their functional and structural constraints shaped over billions of years. Synthetic random sequences, bereft of adaptive significance, lack these signatures and cluster apart in embedding space. By leveraging this evolutionary contrast, the Emory team has devised a method to “peer inside” the black box of AI models, exposing the hallmark features that guide their predictions.

Further validation demonstrated that embeddings flagged as low-quality or uncertain by the random neighbor score tended to perform poorly in downstream biological tasks. These included function prediction, structural modeling, and interaction inference—core applications where misinterpretation can misguide research efforts. Therefore, employing this uncertainty measure not only improves model interpretability but also safeguards the fidelity of scientific conclusions derived from AI predictions, fostering greater trust in computational methodologies.

The simplicity and elegance of the approach belie its broad applicability across the burgeoning landscape of biological language models. As new architectures and training paradigms emerge, providing an intrinsic metric for embedding quality becomes crucial for optimizing model design and application. The concept of a biologically anchored uncertainty score represents a paradigm shift, moving away from generic metrics borrowed from computer science toward domain-specific criteria that align more closely with empirical biological evidence.

Just as a surgeon relies on the sharpest instruments to maximize precision and minimize risk, computational biologists can now choose and refine AI models with enhanced awareness of their limitations and strengths. This precision becomes exceptionally vital when extrapolating to the unknown proteomic “dark matter” found in environmental and clinical microbiomes, where experimental validation lags and computational predictions hold the key to discovery.

By fostering heightened quality control at every stage of protein data modeling—from sequence input through embedding to downstream prediction—this method mitigates the compounding of errors that can arise when working with noisy or unrepresentative datasets. Accurate uncertainty quantification is paramount in a field where even slight errors can propagate through complex biological networks, leading to misleading interpretations or missed opportunities.

This pioneering research was supported by the National Science Foundation and represents a milestone in integrating AI reliability with molecular biology. The work emboldens the interface between computational simulation and empirical biology, highlighting the immense potential of language models while tempering enthusiasm with rigorous validation. As AI-driven biology continues to evolve, transparency and confidence measures like the random neighbor score will shape the trajectory toward more robust and insightful discoveries.

Subject of Research: Not applicable

Article Title: Quantifying uncertainty in protein representations across models and tasks

News Publication Date: 1-Apr-2026

Web References: DOI link

Image Credits: Bromberg lab

Keywords

Bioinformatics, Sequence analysis, Research methods, Complex networks, Computers, Metagenomics, Protein functions

TDFold addresses one of the critical limitations of many existing protein prediction tools: the dependency on multiple sequence alignments and homologous structures. The approach is poised to revolutionize structure prediction by relying solely on individual sequences, bypassing the need for extensive evolutionary data mining. At its core, TDFold utilizes a diffusion model that generates geometrically precise pairwise distance and orientation constraints between amino acid residues in a two-dimensional matrix format. This innovation enables the method to capture spatial relationships inherent to the folded protein without the computational overhead typically associated with homology-based techniques.

The methodology of TDFold is ingeniously split into two primary stages. The first stage involves the generation of two-dimensional geometric templates which effectively capture the pairwise fiber-optic patterns of residue interactions. These templates serve as foundational guides for the subsequent prediction stage. The second stage is characterized by an integrative learning framework where the protein sequence collaborates synergistically with the geometric templates. This collaborative learning paradigm allows for a coherent translation from abstract geometries to detailed three-dimensional conformations, ensuring high fidelity in the final protein model.

One of the groundbreaking elements of TDFold lies in its capacity to outperform state-of-the-art protein language models such as ESMFold and OmegaFold, as well as homology-based giants like AlphaFold2, AlphaFold3, and RoseTTAFold. Notably, TDFold achieves superior predictive accuracy even when working with proteins lacking homologous counterparts, a notorious blind spot for many traditional methods. This capability stems from its innovative use of geometric diffusion mechanisms coupled with a robust training regime that emphasizes generalizability from limited data.

Efficiency in computational biology workflows is paramount, especially for institutions without access to sprawling computational infrastructures. Here, TDFold distinguishes itself not only by its accuracy but also by dramatically reducing resource consumption. The streamlined architecture leverages dimensionality reduction and effective learning strategies to minimize memory usage and inference time. Consequently, researchers in resource-constrained environments—from academic labs to smaller biotech companies—gain unprecedented access to powerful predictive tools that were once dominated by resource-intensive frameworks.

A notable domain where TDFold showcases its prowess is in datasets characterized by homology insufficiency, such as the Orphan and Orphan25 datasets. These datasets comprise protein sequences without clear evolutionary relatives, which traditionally hamper predictive accuracy. Through extensive testing on these challenging datasets, TDFold consistently delivered high-quality structural predictions that rival or exceed those attained by more computationally heavy homologous methods, highlighting its potential in frontier protein engineering and novel protein discovery.

Moreover, the CASP (Critical Assessment of Protein Structure Prediction) benchmarks serve as molten testing grounds for emerging computational models, and TDFold has made significant inroads here as well. In these rigorous and diverse assessments, TDFold has demonstrated its unique balance of speed and precision, often outperforming or matching the predictions of models dependent on substantial evolutionary information. These results underscore TDFold’s potential as a versatile tool that can adapt to a wide range of protein sequences, including those that elude traditional homology-based approaches.

The implications of TDFold are vast. By democratizing access to accurate protein structure models without the prerequisite of large-scale homologous datasets, this technology could accelerate drug discovery, enzyme design, and many other applications dependent on structural insights. The reduction in computational expense and increase in prediction speed imply that previously prohibitive studies can now be undertaken more routinely, expanding the horizon of structural biology research.

From a technical perspective, TDFold’s architecture exemplifies the fusion of machine learning innovations with domain-specific geometric understanding. The two-dimensional diffusion process mirrors physical diffusion phenomena, adjusted to propagate spatial constraints through the residue network. This analogy enables the model to capture complex intra-molecular interactions without resorting to exhaustive combinatorial approaches. Furthermore, the collaborative learning network integrates sequence embeddings with geometric templates, allowing the model to refine its predictions iteratively and coherently.

The success of TDFold also contributes to the growing narrative that single-sequence methods can, under appropriate algorithmic frameworks, challenge the dominance of multiple sequence alignment-based models. This paradigm shift is particularly relevant given the exponential increase in available metagenomic data, where many sequences remain orphaned or await functional characterization. TDFold’s approach ensures that even such sequences can be connected to structural models, providing vital clues about their function and potential utility.

As the scientific community navigates the challenges of understanding the proteome’s vast complexity, tools like TDFold stand out by bridging gaps in dataset availability and computational power. The model’s ability to generalize across protein classes and maintain efficiency will likely inspire a new generation of hybrid predictive frameworks, further accelerating the biological discovery pipeline.

It is also important to highlight the potential educational impact of TDFold. By lowering the barriers to entry for protein structure prediction, universities and teaching institutions can incorporate cutting-edge computational biology into their curricula without the need for substantial computing resources. This democratization of access ultimately fosters a more inclusive scientific environment, nurturing a wider pool of future leaders in protein science.

Despite its promising performance, TDFold also opens avenues for future research and development. Integrating additional modalities such as biochemical constraints, dynamic simulation refinements, or hybrid experimental data could enhance model robustness and applicability to even more challenging protein targets, including large complexes or intrinsically disordered regions.

In conclusion, the introduction of the Two-Dimensional Geometric Template Diffusion method marks a significant leap forward in protein structure prediction. By combining geometric precision, collaborative learning, and efficient resource utilization, TDFold sets a new benchmark for single-sequence-based modeling. This breakthrough not only advances theoretical understanding but also transforms practical approaches, making protein structure prediction faster, more accessible, and exquisitely accurate. Its impact is poised to resonate across the life sciences, from molecular biology research to pharmaceutical innovation.

Subject of Research: Protein structure prediction using single amino acid sequences without reliance on homologous information.

Article Title: Two-dimensional geometric template diffusion for boosting single-sequence protein structure prediction.

Article References:
Wang, X., Zhang, T., Cui, Z. et al. Two-dimensional geometric template diffusion for boosting single-sequence protein structure prediction. Nat Mach Intell (2026). https://doi.org/10.1038/s42256-026-01210-2

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s42256-026-01210-2

Typically, antibodies serve as crucial tools in both biology and medicine due to their highly selective nature, enabling them to bind precisely to specific target molecules. However, a substantial limitation of conventional antibodies is their inability to function effectively within the cellular environment. This has hindered the exploration and understanding of vital biological processes that occur inside cells. Intrabodies present a promising solution to this dilemma, as they are designed to operate within living cells. Yet, their development has historically faced challenges, particularly as antibodies frequently misfold or lose their functional capabilities when introduced into cellular settings.

The research team, led by Professor Hiroshi Kimura from the Institute of Integrated Research at the Institute of Science Tokyo, Japan, collaborated with esteemed colleagues from various institutions, including Colorado State University and Kyushu University. Their collaborative efforts culminated in a breakthrough methodology published in the esteemed journal Science Advances. The core of their advancement lies in a unique design strategy that intelligently alters the structural framework of antibodies while preserving the integrity of their antigen-binding regions.

The AI-driven pipeline developed by this research team keeps the critically important target-binding regions intact, while judiciously modifying the surrounding framework. This approach ensures that the resulting intrabodies are not only able to fold correctly but also maintain their stability inside the cellular milieu without sacrificing their binding specificity. The team meticulously tested approximately 26 previously established antibody sequences, with a remarkable success rate: 19 of those were successfully repurposed into functional intrabodies. Significantly, 18 out of the 19 had earlier been ineffective as intrabodies when conventional methods were employed.

This remarkable outcome underscores a pivotal revelation in the realm of protein design; many antibodies that were previously thought to be unfit for intracellular application can be rendered functional through innovative AI-guided redesign strategies. The transformative potential of artificial intelligence in optimizing antibody structures within cellular environments not only enhances their efficacy but also opens avenues for a more nuanced understanding of cellular dynamics as influenced by these intrabodies.

One of the primary focuses of the study was intrabodies specifically targeting modifications in histone proteins—vital components that play indispensable roles in DNA packaging and gene regulation. These histone modifications serve as crucial markers, providing insights into gene activity. However, traditional labeling techniques to study these modifications often fall short, leaving researchers with gaps in their knowledge. The newly designed intrabodies possess the ability to accurately detect and report on these histone modifications, responding dynamically to changes within cellular environments, thereby illuminating the complexities of gene regulation in real-time.

Further validation of the redesigned intrabodies confirmed their remarkable stability and functionality within living cells. The experimental results demonstrated not only their solubility but also their high specificity for target molecules, which is imperative for accurate biological research. The redesigned molecules exhibited consistent and predictable behavior even under varying cellular conditions, showcasing the reliability of this new research tool. Such characteristics are critical for advancing our understanding of myriad biological functions regulated by histone modifications and other intracellular processes.

The implications of this research extend far beyond basic science. By harnessing the capabilities of AI, the researchers aim to streamline the development of intrabodies, making it a faster, cheaper, and more accessible process for researchers worldwide. As the repository of antibody sequence data continues to grow, the potential to convert existing antibodies into functional intracellular probes could revolutionize diagnostics, fluorescence imaging, and therapeutic strategies in various biomedical applications. The implications for clinical research are profound, suggesting that this AI-driven approach could become a cornerstone of future therapeutic research, enabling scientists to tackle complex disease mechanisms with unprecedented precision.

In conclusion, this innovative study demonstrates that merging artificial intelligence with the nuanced understanding of protein engineering yields a powerful combination for biomolecular research. The team has set a new gold standard for intrabody development, showcasing the transformative potential of AI technologies in modern biotechnology. This research not only provides the foundation for future advancements in the field but also emphasizes the critical role of interdisciplinary collaboration in driving scientific progress forward.

The path ahead is laden with opportunities as researchers look to refine and expand upon these AI-assisted strategies, ultimately aiming to elucidate the intricate networks of biological interactions that govern cellular function and disease processes. By continuing to leverage the capabilities offered by AI and machine learning, the scientific community is poised to unlock new dimensions in the realm of protein design and intracellular research, forging ahead toward a future marked by innovation and discovery.

Subject of Research: Cells
Article Title: AI-assisted protein design to rapidly convert antibody sequences to intrabodies targeting diverse peptides and histone modifications
News Publication Date: 2-Jan-2026
Web References: Science Advances
References: DOI: 10.1126/sciadv.adx8352
Image Credits: Institute of Science Tokyo (Science Tokyo)

Keywords

Health and medicine, Biomedical engineering, Artificial intelligence, Immunology, Antibodies

Proteins serve as the fundamental machinery behind virtually all biological processes, translating genetic code into functional molecules capable of executing complex tasks. Traditionally, understanding protein function has relied heavily on sequence similarity; however, sequence alone often fails to illuminate deeper functional relationships, especially among distantly related species where sequences have diverged substantially. Here, the research team exploited advancements in structural genomics to bridge this gap, unveiling a comprehensive structural landscape underpinning insect biodiversity.

Central to the study was the reconstruction of a highly resolved phylogenetic tree comprising 4,854 insect species. Spanning all extant orders, this phylogeny acts as a scaffold for subsequent structural analyses, enabling evolutionary insights that incorporate lineage-specific diversifications. The team meticulously curated representative species to generate a massive dataset of 13.29 million protein structure predictions, an achievement unprecedented in scale and scope. Remarkably, 11.63 million of these structures were newly predicted for this study, highlighting the magnitude of previously uncharted molecular territory.

Beyond sheer numbers, the research underscores the power of structure-based clustering approaches. By focusing on three-dimensional conformations rather than linear sequence similarities, the investigators illuminated functional relationships obscured by extensive sequence divergence. Proteins displaying divergent sequences yet maintaining homologous structural architectures were effectively grouped, enabling a fresh perspective on protein families and their evolutionary trajectories. This structural convergence approach facilitated the annotation of an astonishing 7.61 million insect proteins, significantly expanding the functional catalog, including identifying functions for nearly 14% of proteins that had remained previously uncharacterized.

The strategic use of known proteins with well-characterized functions served as queries for structural similarity searches throughout the insect protein universe. This tactic bridged a critical gap in functional genomics, especially given the frequent inadequacy of sequence-based annotation for diverse and rapidly evolving insect proteins. The elucidation of nearly three-quarters of a billion “remote homologs” — proteins related by structure but showing scant sequence similarity— speaks volumes to the extent of undiscovered functional diversity maintained by evolutionary pressures.

One of the study’s most compelling revelations revolves around the cGAS-like receptors (cGLRs), a family integral to innate immunity and antiviral defenses. Despite the vast sequence divergence over hundreds of millions of years of insect evolution, these receptors retained striking structural conservation across all 824 representative insects included in the atlas. This finding not only highlights the power of structural genomics to uncover functionally critical proteins missed by sequence analyses alone, but also hints at deeply conserved molecular mechanisms underlying immune defense across insects.

Functional assays provided concrete experimental validation by demonstrating that these structurally defined cGLRs actively participate in antiviral signaling pathways in the yellow fever mosquito, a notorious vector of viral pathogens affecting human populations. This discovery opens exciting possibilities for vector biology and vector control strategies, potentially unveiling new molecular targets to disrupt pathogen transmission by mosquitoes. Importantly, it underscores how structural studies can translate into mechanistic insights with real-world biomedical implications.

Taken together, the integration of large-scale phylogenetic reconstruction with structural predictions marks a transformative shift in molecular biology. Rather than relying solely on traditional sequence comparisons, this framework leverages three-dimensional protein landscapes to chart evolutionary and functional relationships crossing vast biological timescales. It brings to light the evolutionary persistence of crucial protein structures that transcend the limitations imposed by sequence evolution.

The research team’s contributions not only fill significant gaps in our understanding of insect biology but considerably advance the field of structural genomics. By providing an open, richly annotated protein structure atlas, future studies across diverse disciplines—from entomology to immunology and evolutionary biology—stand to gain unprecedented access to molecular blueprints bridging genotype and phenotype.

Moreover, the study’s methodological innovations illustrate the growing importance of integrating computational predictions with experimental validation. The vast majority of this structural atlas was inferred through cutting-edge computational algorithms, yet the concrete experimental verification of cGLRs establishes a model for translating structural annotations into biological functions. This synergy between in silico and in vivo investigations augurs well for accelerated discoveries.

It is also notable that this work highlights insects: not merely as subjects of ecological or agricultural concern, but as molecular troves harboring evolutionary secrets encoded in protein architectures. The staggering diversity of insect species, coupled with their varied ecological niches, suggests a treasure trove of unique proteins shaped by natural selection to fulfill specialized roles. Mining this diversity through the lens of structural genomics opens pathways to novel biomolecules with potential biotechnological applications.

In a broader biological context, such comprehensive structural explorations redefine the concept of homology. Where sequences falter in illuminating distant evolutionary relationships, structural conservation emerges as a robust criterion. This advance has profound implications for annotating unknown proteins across myriad species, potentially unraveling the molecular fabric of life’s tree far beyond insects.

As computational power and protein structure prediction methods continue improving, future iterations of such atlases may expand to other taxa, integrating functional genomics data, transcriptomics, and metabolomics to contextualize structural data within entire biological systems. The landscape painted by this pioneering research sets a new benchmark, showcasing how big data, phylogenetics, and structural biology can converge to shine light on the origins and workings of life’s most complex molecular machines.

This study stands as a testament to the transformative potential of structural genomics in unraveling nature’s molecular secrets. By unveiling an expansive protein structure atlas aligned with insect evolutionary history, it provides a cornerstone for exploring protein function, evolution, and diversity across the most species-rich animal lineage on the planet. With such profound implications for basic biology and applied sciences alike, this research heralds a new era in which three-dimensional structures unlock the mysteries hidden within the genomes of life’s myriad forms.

Subject of Research: Protein structure and function relationships across insect biodiversity.

Article Title: Structural genomics sheds light on protein functions and remote homologs across the insect tree of life.

Article References:
Wu, W., Cui, C., Zhu, Y. et al. Structural genomics sheds light on protein functions and remote homologs across the insect tree of life. Cell Res (2026). https://doi.org/10.1038/s41422-026-01220-0

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s41422-026-01220-0

In recent years, AI techniques have revolutionized structural biology by accurately predicting protein folds from amino acid sequences. Programs like AlphaFold and RosettaFold have transformed our capacity to visualize proteins in silico. These models leverage deep learning architectures trained on known protein structures to infer the spatial arrangements of new protein sequences. The pathbreaking achievements of these methods were heralded by the 2024 Nobel Prize in Chemistry, underscoring their scientific and medical significance.

Yet, despite this remarkable progress, important questions about the underlying mechanisms by which these AI models operate remain unanswered. The latest iterations of these algorithms extend their capabilities beyond predicting isolated protein structures. They now also model how proteins interact with other molecules—commonly referred to as ligands—such as candidate drug compounds. This co-folding or docking prediction holds immense potential for drug discovery, providing a computational shortcut to designing molecules that fit precisely into protein binding sites.

Professor Markus Lill and his team at the University of Basel’s Department of Pharmaceutical Sciences have recently investigated these promising developments with a critical eye. Their research focuses on designing active pharmaceutical ingredients, and naturally, they wondered if current AI models genuinely comprehend the physical chemistry underlying protein-ligand interactions. Given the relatively small dataset of approximately 100,000 protein-ligand structures available for training, they suspected these AI systems might be leveraging superficial pattern recognition rather than fundamental scientific principles.

In their study, Lill’s group introduced deliberate modifications to hundreds of protein sequences to disrupt or alter the chemistry of known ligand binding sites. These included changing the charge distribution dramatically or completely occluding binding pockets. Surprisingly, despite these profound alterations that would normally abrogate ligand binding in reality, the AI models continued predicting the original protein-ligand complex structures, almost as if the modifications had never occurred.

A similar approach was taken with the ligands themselves. By altering the chemical structures of the ligands to prevent any possible interaction with their target proteins, the researchers found that the AI predictions remained largely unchanged. In over half the cases examined, the models failed to account for these perturbations, predicting stable complexes that physically and chemically should not exist.

This compelling evidence suggests that the AI co-folding models do not yet internalize the physicochemical laws that govern molecular recognition and binding affinity. Instead, they appear to rely heavily on data memorization—pattern matching gleaned from their training sets—without a genuine mechanistic understanding. The models can generate plausible-looking structures but lack the capacity to predict outcomes when confronted with novel or deliberately modified molecules and protein sites.

Compounding this issue is the fact that these AI systems struggle considerably when dealing with proteins unlike any they were trained on. When encountering entirely new folds or ligands with no close analogs in the training data, their predictive accuracy drastically decreases. This limitation is particularly consequential given that novel drug targets often involve previously uncharacterized proteins. The inability of these models to generalize beyond known data restricts their utility for cutting-edge drug discovery.

Professor Lill emphasizes a note of caution for the pharmaceutical community. While AI-derived structural models hold great promise for accelerating drug development, relying solely on these predictions without experimental validation or supplementary computational techniques that incorporate physical chemistry can lead to misleading conclusions. Empirical validation remains indispensable to verify AI-based hypotheses and refine candidate drug molecules accordingly.

Looking forward, the researchers propose an exciting direction: integrating the fundamental principles of physics and chemistry directly into AI frameworks. By embedding these constraints and mechanistic insights into machine learning architectures, future models could generate predictions grounded in molecular reality rather than solely on statistical correlation. Such hybrid approaches may yield more accurate and reliable structures, even for uncharted protein-ligand systems.

This integration could profoundly impact drug discovery pipelines by enabling the targeted design of molecules for proteins currently deemed “undruggable” due to their complex or elusive structures. Moreover, enhanced model reliability could shorten development timelines, reduce costly experimental iterations, and catalyze novel therapeutic strategies. The union of AI sophistication with physical law could mark the next transformative leap in biomedical research.

The current study, published in Nature Communications, serves as a vital wake-up call highlighting both the dazzling potential and existing shortcomings of AI in structural biology and pharmacology. It underscores the imperative for multidisciplinary approaches that combine machine learning prowess with rigorous physicochemical understanding. This synergy will be essential to unlock the full promise of AI-guided drug design and realize truly transformative healthcare innovations.

As the field races forward, nuanced scrutiny such as that by Professor Lill and colleagues will ensure that AI tools evolve not only in accuracy but in conceptual depth. Such progress will empower researchers to wield AI models as dependable scientific instruments rather than black boxes, ultimately expediting the discovery of life-saving medicines.

Subject of Research:
Physics-based evaluation of deep learning models predicting protein-ligand co-folding structures

Article Title:
Investigating whether deep learning models for co-folding learn the physics of protein-ligand interactions

News Publication Date:
6-Oct-2025

Web References:
https://doi.org/10.1038/s41467-025-63947-5

Keywords

Protein Structure Prediction, AI in Drug Discovery, AlphaFold, RosettaFold, Protein-Ligand Interactions, Deep Learning, Structural Biology, Computational Chemistry, Molecular Docking, Physicochemical Properties, Machine Learning Limitations, Pharmaceutical Sciences

Until recently, predicting the three-dimensional structure of a protein from its amino acid sequence—a phenomenon known as the protein folding problem—remained an elusive challenge for scientists. Traditional methods, such as X-ray crystallography and nuclear magnetic resonance (NMR) spectroscopy, involve cumbersome experimental procedures and extensive time investments, making them less feasible for rapid discoveries. However, machine learning has emerged as a powerful alternative, notably reducing the time and costs associated with protein structure prediction.

At the core of the study is the use of sparse denoising models, which are a class of advanced neural networks designed to effectively generate high-quality protein structures with minimal input noise. These models leverage vast datasets of known protein structures and sequences, learning intricate patterns that inform the folding process. The methodology involves training the model on a myriad of protein data, allowing it to grasp the complex relationships between amino acid configurations and their ensuing three-dimensional shapes.

Sparse denoising models refine this approach by focusing on the most relevant features of the data, effectively filtering out extraneous noise. This efficiency not only expedites the generation of protein structures but also enhances the accuracy of these predictions. The authors demonstrate that their models outperform traditional prediction algorithms, yielding results that are not only faster but notably more reliable. This advancement signals a paradigm shift in how researchers can approach the complexities of protein structure determination.

The implications of this breakthrough extend far beyond mere academic curiosity. Accelerated protein structure generation holds immense potential for various fields, including medicine, biotechnology, and environmental science. For instance, in drug development, the ability to swiftly predict protein structures can significantly shorten the timeline from discovery to market. Pharmaceutical companies could harness this technology to identify new drug targets and optimize existing treatments, paving the way for more effective therapeutic interventions.

Additionally, this research opens doors for new discoveries in synthetic biology—the design of organisms that produce useful substances or perform specific functions. By accurately modeling protein structures, scientists can engineer novel proteins with desired properties, which could lead to advancements in biofuels, materials science, and agricultural productivity. The capacity to customize protein functions could rewrite the rules for how biological systems are designed and manipulated.

Despite the excitement surrounding these findings, researchers emphasize the need for caution. While their models demonstrate remarkable proficiency, they recognize that no computational method is foolproof. The complexity of biological systems means that unexpected interactions or conformations may still arise, emphasizing the importance of continued experimental validation. Thus, merging computational predictions with laboratory experiments will be paramount to ensure robust outcomes in practical applications.

Moreover, the study highlights the necessity for an interdisciplinary approach in the field of protein research. Collaboration between computer scientists, biologists, and chemists will be instrumental in refining these models and expanding their applicability. By pooling insights and techniques from diverse scientific disciplines, the journey toward a more comprehensive understanding of protein behavior and interaction can be greatly accelerated.

As we reflect on the significance of this work, it is evident that the landscape of protein research is evolving. The advent of sparse denoising models illustrates how artificial intelligence can complement traditional biological research, offering new avenues for exploration and discovery. In a world where the complexities of life continue to challenge our understanding, innovative tools like these inspire hope and curiosity, urging scientists to push the boundaries of what is possible.

The research encapsulates an exciting frontier in computational biology, one where the synergy between machine learning and life sciences continues to flourish. As technology progresses and our understanding of protein structures deepens, we are not just witnessing a scientific evolution; we are participating in a revolution that may one day unlock the mysteries of life itself. The capacity to design, predict, and replicate proteins at unprecedented speeds could transform our approach to disease treatment, environmental challenges, and the sustainable production of goods.

In summary, Jendrusch and Korbel’s work is not merely a technical achievement; it represents a new era in how we approach the structure and function of proteins. The efficiency afforded by sparse denoising models promises to bridge gaps within the scientific community, fostering collaboration and innovation. As more researchers adopt these cutting-edge techniques, the potential for groundbreaking discoveries is limitless, ushering in a future where understanding life at the molecular level becomes increasingly attainable.

The outcomes of this study could very well define the next chapter in protein research, characterized by swift advancements and unprecedented insights. As we anticipate further developments in this area, the scientific community remains poised for a wave of exploration that could reshape our understanding of biology and its applications markedly.

With the publication of this study, we stand at the intersection of technology and biology, waiting to see how these advancements will redefine the nature of molecular research. The journey towards unraveling the complexities of life continues, with sparse denoising models paving the way for a future rich in possibilities.

Subject of Research: Protein Structure Generation with Sparse Denoising Models

Article Title: Efficient Protein Structure Generation with Sparse Denoising Models

Article References:

Jendrusch, M., Korbel, J.O. Efficient protein structure generation with sparse denoising models.
Nat Mach Intell 7, 1429–1445 (2025). https://doi.org/10.1038/s42256-025-01100-z

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s42256-025-01100-z

Keywords: Sparse Denoising Models, Protein Structure, Machine Learning, Computational Biology, Protein Folding

Developed by researchers at EPFL’s Laboratory of Protein Design and Immunoengineering (LPDI), with key collaboration from experts at MIT, BindCraft upends the conventional protein design pipeline. Instead of inputting amino acid sequences into neural networks and then laboriously testing thousands of candidates for proper binding, BindCraft begins with desired structural properties. Using Google DeepMind’s AlphaFold2, it reverse-engineers new protein binders tailored to specific targets. This approach dramatically reduces the requisite trial numbers from tens of thousands to just a handful, enabling labs with limited computational experience to engage in binder design effectively.

At the core of BindCraft is the innovative use of protein structure prediction networks not merely as passive predictive tools but as active generative engines. By harnessing AlphaFold2’s ability to predict protein folding with remarkable accuracy from given sequences, the team ingeniously flipped the process, generating sequences that yield the desired folded shape and functional interactions from the outset. This “reverse engineering” circumvents traditional randomness and uncertainty, focusing computational power on generating binders that inherently possess needed functional features.

The efficiency gains from this strategy are significant. Conventional high-throughput screening approaches require vast experimental and computational resources to sift through enormous combinatorial spaces of protein variants. BindCraft’s design philosophy targets quality over quantity, drastically lowering the barrier to entry for rational protein engineering. By conducting focused screening on a small pool of computationally optimized candidates, researchers can iterate faster and with higher confidence in therapeutic applicability.

To validate their approach, the LPDI team selected a diverse panel of biologically and clinically relevant target molecules. These included adeno-associated viruses (AAVs), commonly employed vectors for gene delivery; the CRISPR-Cas9 system, a revolutionary gene-editing nuclease; and various prevalent allergens. Across this spectrum, BindCraft’s generated binders demonstrated a striking average success rate of 46% in binding assays. This performance indicates not only functional specificity but also the potential to finely modulate protein interactions for therapeutic advantage.

Specifically, in the context of AAVs, the novel binders are envisioned to direct gene therapy vectors precisely to intended tissues or cell types while minimizing off-target effects that have historically posed risks. In gene editing applications, such as those involving CRISPR-Cas9, the binders can serve as regulatory checkpoints, halting the nuclease’s activity spatially or temporally to increase editing safety profiles. This degree of control over therapeutic modalities represents a paradigm shift, enhancing both efficacy and precision.

The impact of BindCraft has been swift and broad. Since its initial preprint release, the platform has garnered extensive attention from academic and industry groups alike. Scientists worldwide rapidly adopted the tool, indicating a strong demand for accessible, user-friendly protein design methods. This enthusiastic reception has also generated valuable user feedback, prompting the development team to expand the platform’s capabilities further, including adapting the method for challenging targets like smaller peptides and other therapeutically promising molecules.

BindCraft’s integration of deep learning with structural biology affords it a unique position at the frontier of bioengineering. Unlike black-box machine learning models that operate solely on sequence data, BindCraft leverages three-dimensional structural predictions directly, ensuring generated proteins meet stringent folding and interaction criteria. This structural awareness greatly enhances the functional viability of designed binders, reducing downstream optimization headaches and accelerating translational research.

Beyond practical applications, BindCraft also contributes to the fundamental understanding of protein interaction landscapes. By successfully designing novel binding proteins that operate at an atomic level precision, researchers can dissect interaction motifs and energy landscapes more rigorously. This improved comprehension enhances predictive models and rational design efforts, benefiting fields across molecular biology, immunology, and synthetic biology.

The BindCraft methodology also heralds increased democratization in protein engineering. Historically, high computational demand and the need for specialist expertise limited protein design to elite laboratories. BindCraft’s streamlined pipeline reduces these barriers, opening doors for a broad array of research groups and fostering innovation diversity. The potential ripple effects in therapeutics development, from precision immunotherapies to gene therapies, are profound.

Importantly, while early results are promising, the field will benefit from comprehensive benchmarking against traditional techniques and longitudinal studies assessing binder stability and in vivo efficacy. The multidisciplinary nature of BindCraft, straddling computer science, structural biology, and molecular engineering, underscores the increasing necessity for collaborative efforts to harness AI-powered biodesign fully. EPFL’s initiative exemplifies how strategic coupling of cutting-edge neural networks like AlphaFold2 with domain expertise can create transformative research tools.

Looking forward, the LPDI team’s plans to extend BindCraft’s capabilities foreshadow exciting advances. Adapting the platform for smaller molecular targets such as peptides paves the way for designing next-generation molecules that may leverage enhanced cell penetration, stability, or multifunctionality. Continued integration of AI with experimental validations will likely yield a virtuous cycle of model refinement and breakthrough innovations in protein therapy.

In summary, BindCraft addresses a pivotal need in molecular biotechnology by making bespoke protein binder design more precise, efficient, and accessible. Its innovative use of protein structure prediction networks to generate candidate sequences represents a conceptual leap beyond traditional trial-and-error methodologies. By democratizing design capabilities and accelerating therapeutic discovery, BindCraft stands poised to significantly impact how drugs and molecular tools are developed in the coming era of synthetic biology.

Subject of Research: Protein engineering and design of functional protein binders

Article Title: BindCraft: one-shot design of functional protein binders

News Publication Date: 27-Aug-2025

Web References:
https://www.nature.com/articles/s41586-025-09429-6

References:
BindCraft: one-shot design of functional protein binders, Nature, DOI: 10.1038/s41586-025-09429-6

Image Credits: 2025 LPDI EPFL CC BY SA

Keywords: Proteins, Biotechnology, Machine learning, Protein folding, Medical treatments, Gene editing