Programming biological function—designing bespoke proteins to perform specified molecular tasks—is a foundational goal of molecular engineering. However, current design models remain fundamentally limited1–3: they typically require either natural proteins as starting points for optimization or manual reformulation of functional goals as geometric and sequence-level constraints to guide candidate generation. Here we introduce Pinal, a 16-billion-parameter foundation model that designs candidate proteins from natural-language descriptions of desired function. Trained on 1.7 billion synthetically annotated protein–text pairs, Pinal links functional intent to protein sequence and structure. In computational evaluations, generated candidates combined high predicted foldability with functional-description alignment and sequence diversity, providing a basis for prioritizing experimentally testable designs. We applied Pinal to four distinct design tasks—a fluorescent protein, a polyethylene terephthalate hydrolase, an alcohol dehydrogenase and a metabolic H-protein—and experimentally observed the intended function in each case, including catalytic activity for both designed enzymes. Without iterative experimental optimization, a Pinal-designed H-protein increased product titre by 1.7-fold relative to the corresponding native Escherichia coli H-protein in a multi-enzyme CO2 fixation pathway. These findings support natural language as a high-level interface for candidate generation in protein design, enabling programmable exploration with reduced reliance on manually specified structural or sequence constraints. Pinal, a 16-billion-parameter artificial intelligence model, designs functional proteins from natural-language descriptions, demonstrating experimentally validated activities and improved CO2-fixation product yields without iterative optimization.

Koch et al. propose replacing “synthetic microbial communities (SynComs)” with “defined microbial communities”. We argue that this substitution obscures the design-driven logic of SynCom research, and that the field needs clearer conceptual classification rather than a broader renaming of SynComs.

Ordering DNA for AI-designed toxins doesn’t always raise red flags.

Amid growing debates about the benefits and risks of studying looking-glass versions of life’s building blocks, there is an urgent need to bridge divergent views.

Mitochondria fulfill central functions in metabolism and energy supply. They express their own genome, which encodes key subunits of the oxidative phosphorylation system. However, central mechanisms underlying mitochondrial gene expression remain enigmatic. A lack of suitable technologies to target mitochondrial protein synthesis in cells has limited experimental access. Here, we silenced the translation of specific mitochondrial mRNAs in living human cells by delivering synthetic peptide-morpholino chimeras. This approach allowed us to perform a comprehensive temporal monitoring of cellular responses. Our study provides insights into mitochondrial translation, its integration into cellular physiology, and provides a strategy to address mitochondrial gene expression in living cells. The approach can potentially be used to analyze mechanisms and pathophysiology of mitochondrial gene expression in a range of cellular model systems.

Machine learning has supercharged the field of computational protein design.

Digital art techniques can now devise custom, working biomolecules on demand.

The structures of engineered odorant receptors.

Microbiome research is now demonstrating a growing number of bacterial strains and genes that affect our health1. Although CRISPR-derived tools have shown great success in editing disease-driving genes in human cells2, we currently lack the tools to achieve comparable success for bacterial targets in situ. Here we engineer a phage-derived particle to deliver a base editor and modify Escherichia coli colonizing the mouse gut. Editing of a β-lactamase gene in a model E. coli strain resulted in a median editing efficiency of 93% of the target bacterial population with a single dose. Edited bacteria were stably maintained in the mouse gut for at least 42 days following treatment. This was achieved using a non-replicative DNA vector, preventing maintenance and dissemination of the payload. We then leveraged this approach to edit several genes of therapeutic relevance in E. coli and Klebsiella pneumoniae strains in vitro and demonstrate in situ editing of a gene involved in the production of curli in a pathogenic E. coli strain. Our work demonstrates the feasibility of modifying bacteria directly in the gut, offering a new avenue to investigate the function of bacterial genes and opening the door to the design of new microbiome-targeted therapies. Edited bacteria were stably maintained in mouse gut for at least 42 days following the delivery of a base editor using an engineered phage-derived particle to modify Escherichia coli colonizing the gut.

Evolution guides biological systems to populate ecological niches, with viruses among the most successful examples of this principle. Viruses evolved over billions of years to efficiently transfer genetic information. Although viruses are highly diverse, most have converged towards remarkable similarity in the size and shape of their capsids1,2. By contrast, generative models for protein design enable the creation of protein architectures that are absent from nature3–5. Here we investigate whether protein assemblies designed by artificial intelligence can be functionalized to construct nucleic acid transport vehicles that are independent of evolutionary trajectories. By combining natural protein domains with synthetic protein assemblies, we create more than 100 bottom-up RNA transfer vehicles with unique sizes and shapes. These vehicles surpass the RNA transfer efficiency of widely used delivery vehicles by several orders of magnitude. In addition, we demonstrate that their tropism can be programmed by incorporation of computationally designed peptide binders and use them to deliver therapeutically relevant cargo RNAs into a wide range of cellular models. We show the in vivo biodistribution of one of these vehicles in a mouse at near-single-cell resolution, confirm its safety, and use it to perform a gene-editing treatment strategy for Duchenne muscular dystrophy in patient-derived cells and a pig. Our work demonstrates how proteins created by generative artificial intelligence can be harnessed for the rational engineering of RNA transport systems with the desired properties by overcoming the limitations of natural protein diversity. Synthetic virus-like protein architectures designed by artificial intelligence show superior ability to deliver RNA into cells, compared with their naturally occurring counterparts, by overcoming the evolutionary constraints limiting viral evolution.

A nanodevice inspired by bacterial contractile injection systems enables reversible, fuel-actuated membrane penetration and spatio-temporally controlled cargo delivery for programmable biochemical regulation in synthetic cells.

Scientists used AI to write coherent viral genomes, using them to synthesize bacteriophages capable of killing resistant strains of bacteria.

Global energy production can't keep pace with AI's demands

Advanced gene editing methods have accelerated biomedical discovery and hold great therapeutic promise, but safe and efficient delivery of gene editors remains challenging. In this study, we present a virus-like particle (VLP) system featuring nucleocytosolic shuttling vehicles that retrieve pre-assembled Cas-effectors via aptamer-tagged guide RNAs. This approach ensures preferential loading of fully assembled editor ribonucleoproteins (RNPs) and enhances the efficacy of prime editing, base editing, trans-activators, and nuclease activity coupled to homology-directed repair in multiple immortalized, primary, stem cell, and stem-cell-derived cell types. We also achieve additional protection of inherently unstable prime editing guide RNAs (pegRNAs) by shielding the 3′-exposed end with Csy4/Cas6f, further enhancing editing performance. Furthermore, we identify a minimal set of packaging and budding modules that can serve as a platform for bottom-up engineering of enveloped delivery vehicles. Notably, our system demonstrates superior per-VLP editing efficiency in primary T lymphocytes and two mouse models of inherited retinal disease, highlighting its therapeutic potential.

The human genome contains millions of candidate cis-regulatory elements (cCREs) with cell-type-specific activities that shape both health and many disease states1. However, we lack a functional understanding of the sequence features that control the activity and cell-type-specific features of these cCREs. Here we used lentivirus-based massively parallel reporter assays (lentiMPRAs) to test the regulatory activity of more than 680,000 sequences, representing an extensive set of annotated cCREs among three cell types (HepG2, K562 and WTC11), and found that 41.7% of these sequences were active. By testing sequences in both orientations, we find promoters to have strand-orientation biases and their 200-nucleotide cores to function as non-cell-type-specific ‘on switches’ that provide similar expression levels to their associated gene. By contrast, enhancers have weaker orientation biases, but increased tissue-specific characteristics. Utilizing our lentiMPRA data, we develop sequence-based models to predict cCRE function and variant effects with high accuracy, delineate regulatory motifs and model their combinatorial effects. Testing a lentiMPRA library encompassing 60,000 cCREs in all three cell types further identified factors that determine cell-type specificity. Collectively, our work provides an extensive catalogue of functional CREs in three widely used cell lines and showcases how large-scale functional measurements can be used to dissect regulatory grammar. Lentivirus-based reporter assays for 680,000 regulatory sequences from three cell lines coupled to machine-learning models lead to insights into the grammar of cis-regulatory elements.

Emerging protein-design competitions aim to sift out the functional from the fantastical. But researchers hope that the real prize will be a revolution for the field.

Non-ribosomal peptide synthetases are assembly line biosynthetic pathways that are used to produce critical therapeutic drugs and are typically arranged as large multi-domain proteins called megasynthetases1. They synthesize polypeptides using peptidyl carrier proteins that shuttle each amino acid through modular loading, modification and elongation2 steps, and remain challenging to structurally characterize, owing in part to the inherent dynamics of their multi-domain and multi-modular architectures3. Here we have developed site-selective crosslinking probes to conformationally constrain and resolve the interactions between carrier proteins and their partner enzymatic domains4,5. We apply tetrazine click chemistry to trap the condensation of two carrier protein substrates within the active site of the condensation domain that unites the first two modules of tyrocidine biosynthesis and report the high-resolution cryo-EM structure of this complex. Together with the X-ray crystal structure of the first carrier protein crosslinked to its epimerization domain, these structures highlight captured intermodular recognition events and define the processive movement of a carrier protein from one catalytic step to the next. Characterization of these structural relationships remains central to understanding the molecular details of these unique synthetases and critically informs future synthetic biology design of these pathways. Structural studies of tyrocidine synthetase using site-selective crosslinking probes to link condensation domains with carrier protein substrates define key interactions and molecular mechanisms of non-ribosomal protein synthesis.

Cis-regulatory elements (CREs) control gene expression, orchestrating tissue identity, developmental timing and stimulus responses, which collectively define the thousands of unique cell types in the body1–3. While there is great potential for strategically incorporating CREs in therapeutic or biotechnology applications that require tissue specificity, there is no guarantee that an optimal CRE for these intended purposes has arisen naturally. Here we present a platform to engineer and validate synthetic CREs capable of driving gene expression with programmed cell-type specificity. We take advantage of innovations in deep neural network modelling of CRE activity across three cell types, efficient in silico optimization and massively parallel reporter assays to design and empirically test thousands of CREs4–8. Through large-scale in vitro validation, we show that synthetic sequences are more effective at driving cell-type-specific expression in three cell lines compared with natural sequences from the human genome and achieve specificity in analogous tissues when tested in vivo. Synthetic sequences exhibit distinct motif vocabulary associated with activity in the on-target cell type and a simultaneous reduction in the activity of off-target cells. Together, we provide a generalizable framework to prospectively engineer CREs from massively parallel reporter assay models and demonstrate the required literacy to write fit-for-purpose regulatory code. A generalizable framework to prospectively engineer cis-regulatory elements from massively parallel reporter assay models can be used to write fit-for-purpose regulatory code.

For Lucas Farnung, there is no question more fascinating than how a single fertilized egg develops into a fully-functioning human. As a structural biologist, he is studying this process on the smallest scale: the trillions of atoms that must synchronize their work to make it happen.