Artificial intelligence is moving from stand-alone models towards systems of interacting agents. These systems do not merely classify, predict or generate: they plan, exchange messages, call tools, divide labour, adapt and coordinate action across tasks and time. This shift has made agentic AI one of the dominant frames for the next phase of the field. Understanding how these agents interact will therefore be as important as improving what each agent can do alone. Here, the authors argue that agentic AI requires a shift from evaluating individual agents alone to analyzing interaction architectures, proposing a framework that links agent interactions, collective configurations and time-dependent outcomes.

Replication origins are essential for both chromosome duplication and plasmid maintenance; however, their annotation remains challenging across different replicon types and across diverse phylogenetic lineages. DoriC 14.0, the latest update of the database of replication origins in prokaryotic genomes, restructures and greatly expands this resource, with particular emphasis on plasmid replication. The latest release contains 530 178 computationally derived origin records, associated with 432 232 distinct sequence accessions, including 434 060 chromosomal and 96 118 plasmid records from bacteria and archaea. For bacterial plasmids, OriV-Finder was applied with RNA II-based refinement, whereas a customized OriV-Finder pipeline was employed for archaeal plasmids, which relies on archaeal replication-associated proteins and origin recognition box motifs. DoriC 14.0 combines literature evidence with computational predictions, maintains traceable record identifiers, and offers versioned data with release history. New features include comparative sequence support, related-origin links, interactive candidate replication initiation protein (RIP) sequence-similarity networks, plasmid-level analyses of origin recurrence, RIP profiles, and RNA II variation. The updated website provides filter-based browsing, record-level views, BLAST searches, and standardized downloads, and is freely accessible at https://tubic.org/doric/ and https://tubic.tju.edu.cn/doric/ without registration.

Quantitative binding affinities and three-dimensional protein–ligand structures provide complementary information essential for structure-based drug discovery, yet these data remain largely disconnected across existing databases and the biomedical literature. We present BioLiP3, a major extension of BioLiP that expands quantitative affinity coverage 13-fold within its structure-centered framework for biologically relevant interactions. BioLiP3 contains 593 423 affinity records, of which 591 820 are linked to three-dimensional protein–ligand structures: 45 386 experimentally resolved PDB complexes, 528 732 BindingDB-derived records with computationally generated structures, and 17 702 structure-linked records derived from LLM-mined biomedical literature. For records lacking experimental complex structures, we generate structural models using available experimental or predicted receptor structures, binding-site identification, and molecular docking, with experimental and computational structures explicitly distinguished. To complement existing databases, we developed a hybrid literature-mining workflow combining large language models with rule-based validation and independent verification to extract Ki, Kd, IC50, and EC50 measurements from PubMed abstracts while preserving provenance. BioLiP3 further integrates ligand-binding sites and protein, functional, and structural annotations via cross-references to PDB, UniProt, NCBI Taxonomy, PubChem, and original literature. Taken together, these advances make BioLiP3 a unified resource for data-driven protein–ligand interaction studies and structure-based drug design. BioLiP3 is freely accessible at https://zhanggroup.org/BioLiP/.

Bidirectional promoters (BDPs), intergenic regions that coordinate transcription of oppositely oriented genes, pervade eukaryotic genomes, yet sequence signatures remain poorly defined. Existing approaches to BDP identification rely on distance thresholds and cannot infer activity from DNA. Here, we use Evo2, a genomic language model, to predict BDP activity and uncover sequence-encoded regulatory principles across four plant species (Arabidopsis, maize, rice, and wheat). We stratify BDPs by co-expression strength and compare Evo2-derived embeddings against k-mer profiles, transcription factor binding site matrices, and convolutional neural networks trained on sequence. Evo2 embeddings outperform all baselines, demonstrating that genomic language models capture regulatory information underlying BDP activity. Models trained on one or multiple species transfer to held-out species, and in planta validation in Spartina alterniflora confirms predicted high-activity BDPs. Sparse autoencoder analysis links Evo2 embeddings to nucleosome positioning and species-specific transcription factor landscapes, providing insight into sequence determinants underlying BDP activity . This work establishes how genomic language models can extract bidirectional promoter logic from sequence and identify conserved regulatory rules across plants. Bidirectional promoters (BDPs) are intergenic regions that initiate transcription of two oppositely oriented genes. Here, the authors report a framework that employs Evo2 genomic language model embeddings to predict BDP activity directly from raw DNA sequence.

RNA chemical probing generates nucleotide-resolved profiles in vitro and in cells, while X-ray crystallographic or cryo-electron microscopy models provide high-resolution structural models but remain comparatively low throughput. Aligning these experiments from different modalities may provide insights into RNA structure formation and molecular functions. Here, we present Ribocentre-FoldBridge (https://foldbridge.ribocentre.org/), an AI-assisted, curator-reviewed database linking curated chemical-probing profiles with experimentally determined 3D structures. This database catalogues 12 types of RNAs, linking 256 929 probing profiles across five experimental categories to 5321 structure-linked cases. It interactively visualizes the probing profiles that align with RNA secondary and tertiary structures at nucleotide resolution. We demonstrate probe-specific reactivity patterns across paired, unpaired, and solvent-accessible structural contexts. The database aims to provide a reproducible evidence layer for examining RNA structure–function relationships across heterogeneous experiments, and thus facilitate the development of new RNA structure or function prediction programs.

The Evolutionary Classification of Domains (ECOD) organizes protein domains into a hierarchy of evolutionary relationships, encompassing both experimentally determined and computationally predicted structures. ECOD now classifies 2.9M domains, of which ∼1.37M derived from the Protein Data Bank and 1.51M from predicted models in the AlphaFold Protein Structure Database. The principal development we report here is a complete rebuild of the ECOD web resource. Search has been reconstructed around three modalities: (i) sequence, (ii) structure, and (iii) keyword. Foldseek now is the primary search method for the ECOD web page, searching a ∼457k representative domain database that is also distributed for local use. Results are presented through a Mol* interactive viewer, and a new API has been developed for programmatic access to the database. Alongside the interface, the pipeline for classifying new PDB entries has been rebuilt and returned to a regular cadence, family groups have been synchronized with Pfam v38.2, the bulk distributables have been reorganized and now include the search databases, the ligands in contact with each domain are recorded for experimental structures and annotated with drug cross-references from DrugDomain, and a registry has been introduced through which ECOD assigns stable identifiers to gene-called proteins and records the structures it predicts. ECOD is freely available at http://prodata.swmed.edu/ecod.

Gluten, the main storage protein in wheat, barley, and rye, contains sequences rich in proline and glutamine that resist complete gastrointestinal degradation resulting in generation of immunogenic peptides. The peptides contain immunogenic epitopes that, following deamidation by tissue transglutaminase 2 (TG2), can be presented by HLA-DQ2/DQ8 molecules to gluten-specific CD4 + T cells, resulting in intestinal inflammation. This gluten-induced intestinal inflammation is known as celiac disease (CD) in susceptible individuals. A gluten-free diet (GFD) has been recognised as gold standard treatment to date, which requires lifelong strict compliance, therefore demanding alternative novel therapies. Fermented foods have emerged as promising reservoirs for selection of gluten-degrading probiotic strains as adjunctive therapeutic candidates. These strains may hydrolyse gluten and its immunogenic gluten peptides through specialised array of proteolytic enzymatic systems comprising endopeptidases, prolyl endopeptidases, aminopeptidases and dipeptidyl peptidases. This review critically examines the significance of gluten-degrading microorganisms with probiotic potential from fermented foods and their proteolytic enzyme system. Owing to their origin from fermented foods, the effect of fermentation parameters such as pH, strain selection, and substrate composition is also highlighted. The assessment of studies indicates that the fermentation mediated by selected gluten-degrading strains reduces the gluten content; however, these did not detoxify the immunogenic peptides completely. The potential strains support intestinal health through modulation of gut microbiota and attenuation of inflammatory responses which has been validated through in vitro and clinical studies, although their efficacy in managing CD remains unestablished. However, strain-specific differences, incomplete degradation of immunogenic peptides, and the lack of standardised protocols for evaluating gluten detoxification remain significant barriers to their application. 

Watson–Crick base-pairing defines the classical paradigm of nucleic-acid interactions, but growing research reveals noncanonical modes beyond this framework. Here we uncover a mode of interactions by which two DNA strands associate through a ligand-induced conformational rearrangement. Upon binding a small molecule, a structure-switching primary aptamer folds into a conformation that exposes a previously inaccessible docking interface. An independently evolved secondary aptamer recognizes this interface primarily through noncanonical interactions rather than sequence complementarity, reporting the ligand-bound state of the primary aptamer. This interaction is conformation-sensitive: single base-pair changes in the primary aptamer modulate secondary aptamer recognition. Across four chemically distinct small-molecule targets—serotonin, dopamine, testosterone and deoxycorticosterone 21-glucoside—secondary aptamers were identified for three, and the serotonin system was translated into surface-based sandwich assays. These results demonstrate that a ligand-induced conformational state, rather than sequence complementarity or the chemical identity of the ligand, can serve as the epitope for nucleic-acid recognition, expanding the molecular features that DNA can be evolved to detect. Aptamers have generally been understood as nucleic acids that recognize a specific ligand. Now it has been shown that secondary aptamers can recognize the ligand-bound conformation of a primary aptamer, enabling recognition of molecules that lack multiple binding sites for sandwich assays. This mechanism can be integrated into lateral flow assays for targets, including serotonin.

The high redundancy of many bacterial proteomes can be used to evaluate proteome quality and identify sequence errors. We have used MMseqs2 clustering with subsequent filtering to identify clusters that contain sequences from at least 50% of the clustered proteomes to build sets of core proteins that include proteins from 95% of the clustered bacteria. These clusters typically capture >80% of proteins in the bacteria. Because these clusters have highly uniform length (the median cluster has >99% of its proteins at the mode length), short (<75% of mode length) or long (>133%) proteins are likely artifacts. Most “outlier” proteins are found in fewer than 10% of clusters, and “high-outlier” clusters are over-represented in a small fraction of proteomes, which often have poor proteome BUSCO fragment scores. Short-outlier proteins are artifacts; at least 80% of short-outlier genomes contain mode-length copies of the protein, which were missed because of frameshifts, termination codons, or initiation codon choice. MMseqs2 clustering with 50% participation provides robust sets of core bacterial proteins and can be used to identify lower-quality proteomes and proteins.

The gut microbiome plays an important role in maintaining human health. Despite the rapid growth of literature and publicly accessible data on the gut microbiome, a large-scale, well-curated database integrating a unified matrix with multiple analytical tools is still lacking. Here, we introduced GutBiomeDB, a comprehensive database and analysis platform for the human gut microbiome. GutBiomeDB contains 168 464 samples from 482 BioProjects, spanning 72 countries, 224 disease conditions, and seven life stages. To improve the reusability and accessibility of data, the uniformly processed taxonomic profiles generated by Abdill et al. were reused without reprocessing to construct a matrix of 4680 taxa and 3142 genera, and disease, life-stage, sex, and geographic metadata were manually curated by our team. Additionally, GutBiomeDB offers four tools for in-depth analysis: “Differential Analysis” for comparing diversity and abundance of user-selected cohorts and performing random-effects meta-analysis across projects; “Interaction Network” for comparing disease and healthy-control co-occurrence networks; “Lifecycle Atlas” for visualizing genus trajectories across seven life stages; and “Health Index” for searching similar samples and scoring uploaded genus profiles. GutBiomeDB serves as a versatile and valuable resource for advancing a broad spectrum of gut microbiome research and clinical relevance. GutBiomeDB is freely accessed at https://gutbiomedb.online.