Metagenomic databases hold vast quantities of unexplored genetic material that can seed tools for biotechnology and medicine, but discovery is limited by our ability to examine sequence data at scale. Here, we show that large language models (LLMs) in an autonomous harness can alleviate this bottleneck and surface novel molecular systems. We deployed Claude Code instances to survey reverse transcriptase (RT) loci across 1.9 billion protein clusters. Among the top candidates were array-associated RTs (ART), a novel family of jumbo-phage RTs coupled with an array of ~200-nucleotide units and a dedicated partner gene. Session transcripts show that Claude identified the ART family by directly examining its DNA sequences and recognizing the atypical repeats, a behavior attributable to specific Mythos 5 internal signals that respond to repeated DNA. ART arrays are highly expressed and appear as discrete units during Staphylococcus phage infection, suggesting an RT system directed by a repertoire of distinct RNAs. These findings indicate that LLMs can autonomously detect anomalies and drive analyses to initiate biological discoveries.

Protein structural complexity confers responsiveness and modularity, allowing proteins to act as control modules in biomolecular circuitry. Protein engineering that commandeers native cell signaling and central dogma functions has led to the development of cell therapies, specialized biosensors, and a deeper understanding of cellular mechanisms. In this review, we examine how protein control modules command circuit with therapeutic functionality across cellular spatial levels. Extracellularly, signal proteins such as antibodies and cytokines can provide either binary or gradient-sensitive detection of environmental features. Extracellular proteins also serve as inputs to synthetic receptors that transmit signals across membranes, rapidly linking environmental cues to intracellular action. Inside the cell, protein control modules are engineered from central dogma components to bind nucleic acids, creating gene circuits that influence transcription, translation, and nucleic acid processing, and ultimately regulate protein expression. This review covers circuitry platforms and immunoengineering applications across spatial levels, focusing on innovation from 2016 to 2026. With continued development, synthetic biocircuits can serve as a platform for creating next-generation “smart” therapeutics and ultimately for engineering immunity.

Protein inverse folding, which infers compatible amino acid sequences from a given backbone structure, is central to rational protein design. Current methods rely largely on sequence recovery rate as the evaluation metric, emphasizing structural compatibility but seldom addressing function. Yet functional activity depends on the proper network of non-covalent interactions-protein-ligand contacts being a prime example. We argue that function-oriented protein design is governed by three complementary constraints: structural, interaction, and evolutionary, none of which are handled in a unified manner by existing pretrained models. To bridge this gap, we introduce zPolaron, a graph neural network built on protein-ligand complexes for functional protein design. zPolaron represents residues and ligand atoms as a heterogeneous graph and jointly encodes structural, interaction, and evolutionary information by fusing physical-interaction-enhanced graph representations with ESM-2 embeddings. Across multiple fitness prediction benchmarks, zPolaron outperforms all baselines, with gains evident across structural, interaction, and evolutionary dimensions. Wet-lab experiments further confirm its utility in functional enzyme screening and design: the optimized CalA6-D205A-S160G mutant achieves a 17.84-fold improvement in activity, while the CalB-I121M mutant yields a 1.91-fold increase. Together, these results show that integrating interaction and evolutionary information fills the missing puzzle of inverse folding, enhancing protein fitness prediction beyond structural compatibility. zPolaron is available at https://github.com/zelixirSH/zPolaron.

Synthetic biology enables the installation of programable genetic functions in microorganisms, positioning them as versatile platforms for therapeutic applications. However, clinical translation requires chassis organisms that integrate safety, genetic stability, and precise control over the activity. Here, we present an innovative engineering framework for the probiotic E. coli Nissle (EcN) that transforms EcN into a safe, efficient, and non-proliferative chassis for therapeutic production. The native EcN plasmids pM1 and pM2 were cured and repurposed to encode two functional modules: a CRISPR-Cas12 chromosome-shredding device that prevents bacterial propagation and a programable circuit for payload expression, thereby generating a non-proliferative therapeutic-producing system. To broaden the range of therapeutic payloads, we developed an artificial-intelligence-guided bioinformatic pipeline to predict anticancer cell-penetrating peptides (ACCPPs). As a proof of concept, a candidate identified through this workflow was produced in the engineered EcN chassis and evaluated in human cell culture for anticancer activity. Collectively, this strategy established a modular platform that integrates computational therapeutic discovery with engineered probiotic production, paving the way for the future development of safer microbial therapies and their clinical applications.

Rare steviol glycosides (SGs), characterized by superior sensory profiles and promising bioactivities, are high-intensity sweeteners derived from Stevia rebaudiana. Their intrinsically low abundance limits extraction yields, while in vitro bioconversion is limited by the availability of UDP-glucose (UDPG). Here, an E. coli chassis was reprogrammed to achieve an intracellular UDPG content of 11.72 μmol/g CDW, 2.19-fold that of BL21(DE3), by enhancing pyrimidine nucleotide synthesis, blocking competing pathways, redirecting carbon flux, and optimizing the expression of key genes. Coupled with a UGT−SuSy system, the platform enabled in vitro conversion of Reb A into rare derivatives Reb L2, Reb D, Reb D2, Reb M, Reb M2, and Reb M8 without exogenous UDPG, with 1.6- to 129-fold higher conversion efficiencies than the wild type. Cultured in a 5 L bioreactor, harvested cells yielded 108.57 g/L Reb L2 and 111.99 g/L Reb D via in vitro whole-cell biotransformation. This platform provides a route for rare SG production and a versatile foundation for the glycodiversification of natural products.

Prophages are increasingly recognized as drivers of bacterial adaptation, yet how domesticated prophage genes shape microbial symbiosis remains poorly understood. Here, we identify GDPDXhp1, a glycerophosphodiester phosphodiesterase encoded by prophage Xhp1 of Schaalia odontolytica XH001, as a metabolic regulator linking bacterial physiology, episymbiosis and phage fitness. Association with the epibiotic bacterium Nanosynbacter lyticus TM7x strongly induced gdpdXhp1 expression and triggered GDPDXhp1-dependent lipid droplet accumulation in XH001. GDPDXhp1 orchestrated glycerophospholipid remodeling, membrane biophysical properties, respiration and cell-surface glycan composition in XH001, thereby promoting TM7x episymbiosis. Remarkably, GDPDXhp1 was also required for productive Xhp1 infection by remodeling susceptible bacterial host envelope structures to facilitate phage-host interactions. Together, our findings reveal a domesticated prophage enzyme that contributes to bacterial metabolism, interspecies symbiosis and viral reproduction, highlighting prophage as an overlooked source of metabolic innovation in microbial symbiosis.

RNA splicing is frequently disrupted by genetic variation, contributing to a wide range of human diseases. Identifying the variants responsible for altered splicing is therefore crucial but has long been challenging. Deep learning models that predict splicing from genomic sequence have substantially advanced their detection, but accurately resolving their quantitative effects on splicing and the resulting transcripts remains an unmet need. Here we introduce SpliceAI2, a deep learning model trained end-to-end on splice site, splice junction, and transcript usage across multiple mammalian species. SpliceAI2 reconstructs complete transcripts directly from genomic sequence and considerably improves identification of splice-altering variants and quantification of their effects in diverse benchmarks spanning cryptic splicing, splicing quantitative trait loci, and massively parallel reporter assays. Human population datasets provide independent validation of these predicted functional effects through evidence of natural selection and associated changes in plasma protein abundance. Application to a rare disease cohort reveals that predicted splice-altering variants are highly enriched in disease-associated genes and account for 15% of the excess genetic burden, with approximately half residing in deep intronic regions likely missed by exome sequencing. Together, these findings establish SpliceAI2 as a major step forward in interpreting splice-altering variants across the human genome.

In diverse natural communities, cells and phages interact with many potential partners, but the impacts of pairwise coevolution on the overall structure of phage-bacteria interaction networks (PBINs) is not fully understood. Here, we conducted multiple pairwise coevolution experiments and examined how coevolution with one partner could simultaneously influence interactions with other partners. We then analyzed the underlying genetic changes that alter interactions in a large cross-experiment PBIN. The results reveal that interactions at the cell surface as well as intracellular interactions can alter patterns across the PBIN. When phages utilized the same set of host cell-surface receptor genes, pairwise coevolution with one partner did alter interactions with other partners. In contrast, if different sets of host cell-surface receptor genes were used, modular patterns in the cross-experiment PBIN were generally observed. On the phage side, resistance-breaking mutations arose in many putative receptor-binding proteins, and these changes could alter interactions with other hosts. However, in both the hosts and phages, over half of the mutated genes were not associated with cell-surface interactions. Host mutations likely altering intracellular phage-host interactions, including translation, protein stability, and DNA replication often led to broad resistance across the PBIN. In phages, mutations occurring in genes encoding endolysins, DNA methylases, oxygenases, transcriptional regulators, and tRNAs were observed potentially also influencing intracellular interactions. Based on these results, we propose a model to illustrate how these two different types of genetic signatures (i.e. extracellular cell-surface interactions verses intracellular interactions) can impact the structure of PBINs.

Fucoidan is the second most abundant polysaccharide in brown macroalgae, contributing up to 50% of the dissolved organic matter released daily from these algae into the ocean, substantially influencing the marine carbon cycle. Due to its complex chemical structure, fucoidan is highly resistant to microbial degradation and accumulates on the surface of macroalgae, where it contributes to regulating their microbiome. Here, we used a range of microfluidic assays to quantify the chemotactic and physiological responses of marine bacteria to fucoidan. Our results reveal that all the strains tested (14 out of 14) are attracted to fucoidan produced by two different species of brown macroalgae. Attraction occurred only at high fucoidan concentrations (0.1 to 1 mg/mL), such as those expected in the immediate vicinity of algal surfaces, whereas low concentrations elicited no response or modest chemorepulsion. Although fucoidan did not support bacterial growth under our experimental conditions, it did not impair the utilization of other carbon sources. Together, these findings suggest that fucoidan gradients can act as chemical cues guiding marine bacteria toward the nutrient-rich microenvironment of macroalgal surfaces, pointing to a potential ecological role of fucoidan in shaping macroalgal microbiota, bacterial behavior, and ultimately carbon cycling in the ocean.