To advance the computational simulation of cellular life, we propose a virtual yeast, an artificial intelligence (AI)-driven agent that models eukaryotic cellular behaviors by integrating multimodal biological data, mechanistic reasoning and active experimentation using Saccharomyces cerevisiae as a genetically tractable and data-rich model system. Cellular complexity is decomposed into eight function-centred modules, spanning genetic, metabolic and structural systems, each realized as a domain-specific AI tool coordinated through a large language model-based orchestration layer. Built on three data pillars, namely, mechanistic knowledge, subcellular architecture and dynamic states, the system integrates representation learning and generative modelling within a closed-loop learning pipeline that autonomously designs and executes experiments. The virtual yeast serves as both a conceptual and an operational platform to optimize biosynthetic pathways, support the generation and prioritization of hypotheses across diverse cellular processes, and accelerate target discovery. By coupling biological realism with autonomous AI reasoning, the virtual yeast establishes a generalizable blueprint for constructing virtual eukaryotic cells and advancing synthetic biology. A virtual yeast, an artificial intelligence-driven agent for modelling eukaryotic cellular behaviours, is described.

The recent revolution in genome sequencing and protein structure prediction has opened new frontiers in understanding, predicting and designing enzyme function. Central to these efforts is the discovery and functional annotation of novel enzymes, which is essential for elucidating the connection between genotype and phenotype and for developing biocatalysts for industrial applications. However, accurately predicting enzymatic function remains a major challenge, and the discovery of new enzymes often relies on serendipity. Here we present a metal-coordination-guided strategy that uses atomic-level mechanistic principles to mine protein structure databases for the targeted discovery of metalloenzymes. We apply this framework to the AlphaFold2 Protein Structure Database to identify new members of the FeII/α-ketoglutarate-dependent halogenase family, which selectively functionalize unactivated C(sp3)-H-bonds, a crucial transformation in the production of pharmaceuticals and other high-value compounds. These radical halogenases constitute a low-abundance class within the large and diverse cupin superfamily. Owing to low sequence conservation, they have been especially challenging to find against the complex background of related family members, such as hydroxylases, desaturases and epimerases. Our metal-coordination mining methodology reveals several previously unrecognized radical halogenase families spanning diverse phylogenetic space, at minimal computational cost. Our predictions are validated by the experimental characterization of two new radical halogenases, AspX and BtnX. Notably, BtnX shows a substrate promiscuity that is unprecedented in radical halogenases, opening the way for a broad range of biocatalytic applications. A methodology for mining protein structure databases on the basis of the distinct intrinsic structures of metal-binding active sites in enzymes enables the discovery of new families of radical halogenases.

Synthetic mRNA therapeutics offer a versatile platform for treating diverse conditions, including cancer and infectious diseases. For delivery into cells, these mRNAs are encapsulated in lipid nanoparticles and commonly incorporate modified ribonucleotides to improve stability, enhance translation and mitigate immune recognition. N1-Methylpseudouridine (m1Ψ) has become the industry standard for synthetic mRNAs owing to its effectiveness in promoting translation and reducing immunogenicity. However, recent studies have shown that m1Ψ can compromise translational fidelity, leading to errors such as premature termination and ribosomal frameshifting. Here we reveal N4-acetylcytidine (ac4C) as a functionally distinct alternative to m1Ψ. Across cultured cell lines, primary human monocyte-derived dendritic cells and mouse liver, ac4C suppressed inflammatory responses as effectively as m1Ψ while driving higher protein yields. Single-molecule imaging of translation revealed broadly similar ribosome densities per mRNA for ac4C-modified and m1Ψ-modified transcripts. However, translation elongation with m1Ψ-modified mRNA was nearly twofold slower than with ac4C, which resulted in reduced protein output and increased ribosome collisions that further limited protein production through the engagement of quality-control pathways and +1 frameshifting. These findings underscore the importance of context in designing therapeutic mRNAs and position the translation elongation rate as a key determinant of the efficacy of modified ribonucleotides. Different RNA modifications elicit different translation elongation rates for in vitro transcribed mRNAs that result in disparate translation outputs and fidelity, as shown here for N4-acetylcytidine versus the industry standard for synthetic mRNAs, N1-methylpseudouridine.

Prime editing has not been established in filamentous fungi, which are major ecological contributors and industrial hosts with vast biosynthetic capacity. Here we develop fPE7max, a prime editing platform optimized for fungi, which supports different edit types, including base substitutions and defined small insertions or deletions, with an average editing efficiency approaching 90%, across diverse genomic loci and species. fPE7max further enables larger insertions of up to 1 kb and deletions of up to 10 kb. We perturb upstream open reading frames in the pleiotropic regulator gene, laeA, to modulate metabolic output across multiple fungal species. Metabolomic profiling reveals activation of previously lowly biosynthetic pathways, leading to the identification of 18 metabolites, including 8, to our knowledge, previously unreported structures, 3 of which with cytotoxic activity. These results establish fPE7max as an efficient platform for genome engineering in filamentous fungi and show upstream open reading frame editing as a strategy for modulating endogenous regulatory networks and accessing the fungal chemical repertoire. fPE7max efficiently installs base substitutions, insertions and deletions to modulate filamentous fungal metabolism.

Protein-protein interactions (PPI) are central to biological processes. Designing small molecules that modulate dysregulated PPIs holds strong promise for targeting undruggable proteins. However, existing structure-based drug design approaches focus on well-defined small-molecule binding pockets and struggle to generalize to large, shallow, and chemically complex PPI interfaces. Here, we introduce Pep2Mol, a diffusion-based generative model for 3D molecule design that targets orthosteric PPI sites by explicitly incorporating binding peptides or proteins as structural guidance, moving beyond conventional pocket-conditioned generation. To enable model development and benchmarking, we curate a large-scale, high-quality dataset of 10,956 experimentally resolved protein complex structure pairs, each pairing an orthosteric competitive ligand with a protein binder at overlapping receptor interfaces. Pep2Mol integrates two SE(3)-equivariant graph neural networks that encode protein-ligand and protein-peptide interactions respectively, and fuses these representations via attention-based conditioning to jointly guide the diffusion trajectory. Extensive evaluations demonstrate that Pep2Mol generates chemically valid ligands with state-of-the-art binding affinities, providing a strong foundation for small-molecule inhibitor design against challenging PPI interfaces.

Metalloproteins are essential to many cellular processes. They use metal ions as cofactors to catalyze reactions, stabilize protein structures, and mediate electron transfer. Identifying their metal-binding sites remains difficult because of the complexity of protein environments and the promiscuous binding of metal ions, and existing computational methods are limited by accuracy and data scarcity. Here we introduce PRIME, a hybrid deep learning framework that combines evolutionary and structural signals to predict metal-binding sites accurately and efficiently. PRIME employs protein language models and pre-trained structure models to extract information from protein sequences and structures, together with a probe generation algorithm that bridges sequence- and structure-based predictions by scanning candidate sites. PRIME outperforms existing methods across diverse metal ions, from abundant zinc and calcium to challenging potassium and sodium. Ablation analysis shows that pretrained structure models improve accuracy. Case studies on AlphaFold2 models further demonstrate PRIME’s potential for high-throughput metalloproteomics. The Authors present PRIME, a deep learning tool that fuses protein sequence and structure to pinpoint metal-binding sites across diverse metal ions, from abundant zinc to rare potassium, offering improvements in speed and accuracy over existing methods.

The limestone monuments of the Rectangular Tower in the Xiaoling Tomb of the Ming Dynasty, created in the mid-fourteenth century, are biodeteriorating from environmental exposure, resulting in the formation of black biocrusts. However, the microbiomes that shape biocrust formation and the biodeterioration processes involved remain unclear, significantly challenging the conservation of stone monuments at this archaeological site. Here, we systematically investigated the physicochemical properties and microbial communities of biocrusts to identify keystone taxa that shape their formation and biodeterioration. Physicochemical analysis indicated that biological crusts are associated with calcium mobilization and redistribution of the limestone monuments. Microscopy and spectroscopy indicated that microbial interactions with limestone promote the formation of biological crusts. Importantly, we observed the significant predominance of Cyanobacteria and/or Chloroflexi in biocrusts, suggesting that photosynthesis may be a crucial process in biocrust formation. Fungal communities in biocrusts were dominated by Ascomycota, Basidiomycota, and Chytridiomycota, while archaeal communities were dominated solely by Nitrososphaerota. Microbial co-occurrence network and correlation analyses identified 12 keystone taxa across 11 genera that shape biocrust formation. Importantly, Scytonema spp. could provide organic carbon and nitrogen for Spirosomaceae spp., and members of the classes Cyanobacteriia and Agaricomycetes, as well as the genera Setophaeosphaeria and Plectosphaerella, are likely the keystone taxa responsible for both biocrust formation and the associated biodeterioration. Additionally, two predominant ammonia-oxidizing archaeal families (i.e., Nitrososphaeraceae and Candidatus Nitrocosmicus) could support chemolithoautotrophic growth in the microbiome by oxidizing ammonia and fixing carbon dioxide. Together, these findings underscore the need for targeted conservation strategies to mitigate microbial biodeterioration of stone monuments during biocrust formation.

High-throughput screening of protein domains enables the systematic discovery of protein sequences that encode specific cellular functions. Fluorescence-activated cell sorting-based assays have long been the standard readout for such screens but remain time- and resource-intensive, imposing practical limits on library size and coverage. Here we describe a scalable magnetic separation-based workflow that provides an alternative to FACS for screening large protein libraries in mammalian cells. We engineered a modular synthetic surface marker, consisting of a fusion between the fragment crystallizable (Fc) region of human immunoglobulin G and the transmembrane domain of platelet-derived growth factor receptor-β, that allows cells to be magnetically separated on the basis of surface reporter expression using Protein G-coated magnetic beads. The procedure covers pooled library cloning, lentiviral delivery, magnetic separation and sequencing-based quantification, enabling reproducible screening of more than 100,000 protein domain variants. The approach is suitable for the identification of functional protein domains capable of transcriptional and post-transcriptional RNA regulation and may lead to the selection of improved transmembrane domains for efficient protein surface display. The entire workflow, from library design to data analysis, can be completed in 4–6 weeks and requires skills in cell culture, molecular cloning and computational techniques. This scalable and accessible Protocol enables researchers to systematically measure protein domain functions across biological contexts, thus accelerating both biological discovery and protein engineering. This Protocol presents a scalable magnetic separation-based method for screening large protein domain libraries in mammalian cells, providing a simple and quantitative alternative to fluorescence-activated cell sorting for functional protein discovery.