bioRxiv Science⌕ Search

Biology subjects

Xu, Z.

Publications and source records attributed to Xu, Z..

At least 109 records · Page 6Linked to original sources

A self-organizing single-cell morphology circuit optimizes Podophrya collini predatory trap structure

Cellular structure self-organizes through an interplay between internal mechanisms and external cues. In predatory ciliates, structures that capture prey obtain the resources needed for their construction, creating feedback between environmental inputs and morphological outputs. Here, we describe a self-organizing single-cell morphology circuit that adaptively optimizes the predatory trap structure of the suctorian Podophrya collini. These trap structures ensnare large cellular targets using an array of straw-like tentacles that siphon out prey cytoplasm upon contact with their tips. We find that trap architecture scales anisotropically, favoring tentacle number over length, to construct traps that maximize capture probability for the resources on hand. Drug perturbations, transcriptomics, proteomics, and expansion microscopy define distinct molecular and structural requirements that regulate trap structure maintenance and tentacle biogenesis. We integrate these findings into a mathematical model that explains the adaptive scaling of the trap and that makes predictions we confirm experimentally. More broadly, this circuits architecture provides a general-purpose control logic for organizing the number and size of sub-cellular structures applicable to other natural and engineered cellular systems.

cell biology↗

Cold Atmospheric Plasma for Alzheimer's disease

Lightning has been proposed as a pivotal energy source driving prebiotic chemical evolution and early life processes. Similarly, external physical stimuli such as sound and light have been shown to modulate neurophysiological activity and mitigate Alzheimers disease (AD) pathology by promoting amyloid-{beta} (A{beta}) clearance. Here, we demonstrate that cold atmospheric plasma (CAP) acts as a joule-level analog of lightning, integrating light, sound, and reactive species to modulate AD progression in a mouse model. Using multimodal analyses, we show that CAP treatment enhances key neuroimmune signaling pathways in an AD murine model, including microglial activation, without inducing pathological alterations at the functional, transcriptomic, or proteomic levels in healthy mice. These findings highlight CAP as a safe and efficacious modality for modulating neurodegenerative processes, establishing a foundation for its therapeutic translation in AD and related disorders.

bioengineering↗

DEEPReditor-CMG: A deep-learning-based predictive RNA editor for crop mitochondria genomes

C-to-U RNA editing is a pervasive phenomenon in crop mitochondria, playing a pivotal role in the processing of functional proteins encoded by mitochondrial genes and serving as a crucial regulatory mechanism for nuclear control over mitochondrial gene expression. However, identifying editing sites across the mitochondrial genome with high precision remains a formidable challenge. To address this challenge and uncover the potential regulatory patterns of C-to-U RNA editing, this study compiled a comprehensive dataset of mitochondrial genome sequences from 13 crops, including Arabidopsis thaliana, Lactuca sativa var. Capitata, Oryza sativa Japonica, and Zea mays, among others, annotated with editing sites from NCBI. Utilizing TensorFlow 2.0, we constructed a rigorous hyperparameter optimization framework, which facilitated the development of the DEEPReditor-CMG model, demonstrating superior predictive capabilities. Based on the DEEPReditor-CMG model and three interspecies relationship indices, we discovered that species with closer phylogenetic relationships exhibit more similar C-to-U RNA editing mechanisms, providing valuable insights into the evolutionary conservation of RNA editing mechanisms. We further developed executable programs for predicting C-to-U RNA editing sites in crop mitochondrial genomes, leveraging the conserved nature of the crop RNA editing machinery to maximize prediction reliability and uncover new editing sites with great potential. This research not only deepens our understanding of C-to-U RNA editing and post-transcriptional regulatory mechanisms in crop mitochondria but also paves a novel avenue for investigating RNA editing mechanisms. Experimental data, computational code, and applications are publicly accessible at https://github.com/Qinsidong/DEEPReditor-CMG. HighlightsHigh-precision prediction model for C-to-U RNA editing in crops--DEEPReditor-CMG Three interspecies relationship indices describing C to U RNA editing Prediction tools based on the conservation of crop RNA editing mechanisms Open-sourced experimental data, computational code, and applications

bioinformatics↗

Novel Role of AcylCoA:cholesterol acyltransferase 1 (ACAT1/SOAT1) in Diabetic Retinopathy

Hypercholesterolemia and excessive cholesterol ester (CE) production have been linked to chronic inflammation and vascular dysfunction during cardiovascular disease. Upregulation of AcylCoA:cholesterol acyltransferase 1 (ACAT1/SOAT1), the enzyme responsible for retinal CE formation, has been implicated in pathological retinal neovascularization. Here we determine the role of this process in diabetic retinopathy (DR). Ins2Akita diabetic mice were treated with the specific ACAT1/SOAT1 inhibitor K604 (10 mg/Kg, i.p.) beginning at 10 weeks for 2 weeks or 8 months for 2 months. ACAT1/SOAT1 expression and CE formation were assayed along with oxidative stress, inflammation, vascular pathology, and neuronal function. ACAT1/SOAT1 expression was also assayed in human retinas and vitrectomy specimens. Retinas from early-stage Ins2Akita mice exhibited increases in CE deposition, superoxide production, and expression of ACAT1/SOAT1, LDLR, TREM1, MCSF, and VEGF along with leukostasis, vascular leakage, acellular capillary formation, retinal ganglion cell loss, and impaired visual function. Late-stage increases in CE, ACAT1/SOAT, oxidative stress, inflammation, and impaired visual function were also observed. These changes were significantly inhibited by K604 treatment. The protective effects were independent of changes in systemic glucose or body weight. Human retina and vitrectomy samples also showed increases in ACAT1/SOAT1 and CE, respectively. Specific inhibition of ACAT1/SOAT1 with K604 normalizes ACAT1/SOAT1 expression and CE formation and prevents increases in oxidative stress and inflammation and preserves retinal structure and function in both early and late stages of DR. These findings identify ACAT1/SOAT1 as a promising therapeutic target for both early intervention and later stage treatment of DR. One sentence summaryInhibiting cholesterol esterification limits retinal neurovascular injury in diabetes.

cell biology↗

Closed-loop beta stimulation enhances beta activity and motor behaviour

Movement-related beta event-related synchronization (ERS) has been linked to motor control and learning, showing potential as a therapeutic target for those with movement deficits, such as stroke survivors. However, whether directly modulating beta ERS can causally influence motor performance remains unclear, largely due to the lack of methods designed to specifically target this neural activity. To address this gap, we developed a novel behaviourally-driven, closed-loop transcranial alternating current stimulation (tACS) approach to target movement-related beta ERS during a visuomotor adaptation task. We found that the behaviourally-driven, closed-loop beta-tACS specifically enhances beta ERS without affecting beta event-related desynchronization (ERD). Critically, this targeted enhancement significantly improves retention of motor adaptation. These findings establish a causal relationship between beta ERS and motor behaviour and highlight the potential of behaviourally-driven beta-tACS as a therapeutic approach for improving motor function in clinical populations characterized by impaired beta activity.

neuroscience↗

Local translation of circular RNAs is required for synaptic activity and memory

Circular RNAs (circRNAs) are enriched in synapses and implicated in cognitive processes, and recent studies have shown that circRNAs can encode micropeptides, which suggests that there may be novel synaptic proteoforms in the brain that await discovery. Here we report widespread learning-induced local circRNA translation in the prefrontal cortex of male C57BL/6 mice. More than 1500 synapse-enriched circRNAs contain active IRES elements, with 842 interacting with the ribosome and 241 exhibiting direct evidence of activity-induced translation. We discovered a synapse-enriched micropeptide (P1) that is derived from a single exon circRNA, the mRNA host of which encodes an enzyme associated with protein repair. Although P1 is only a third of the size of the full-length protein, it is locally expressed, enzymatically active, and interacts with plasticity-related proteins, including CaMKII. In addition to direct effects on synaptic activity, targeted P1 knockdown impairs whereas its overexpression enhances fear extinction memory. These findings shed new light on the dark proteome in the brain and reveal local circRNA translation as a novel mechanism of plasticity and memory.

neuroscience↗

Meiosis-specific genes play roles in ploidy reduction in Cryptococcus neoformans titan cells

Cryptococcus neoformans is a fungal pathogen of humans that causes life-threatening meningoencephalitis. During infection, enlarged, polyploid titan cells are produced that promote persistence in the host, in part by resisting phagocytosis; under stress conditions, such as exposure to the antifungal drug fluconazole, titan cells can produce aneuploid or diploid, drug-resistant daughter cells. However, the mechanism underlying this ploidy reduction remains poorly understood. Interestingly, meiosis related genes have been shown to be activated during Cryptococcus infection, leading us to hypothesize that the depolyploidization of C. neoformans titan cells may occur through a process resembling the ploidy reduction during meiosis. In this study, we show that titan cells developed from diploid strains predominantly produce diploid daughter cells with haploid daughters observed infrequently. We further demonstrate that meiosis-specific genes, including DMC1 and SPO11, are critical for stable inheritance of a diploid genome in the daughter cells. Specifically, deletion of these genes in a heterozygous diploid background resulted in: 1) titan cells with a significantly reduced capacity to produce daughter cells; 2) increased phenotypic variation among daughter cells produced by the titan cells, including traits that could be relevant to cell growth and viability; and 3) daughter cells produced by the titan cells exhibiting high levels of loss of heterozygosity (LOH) and aneuploidy, suggesting elevated genome instability. Taken together, these findings demonstrate the importance of meiosis-specific genes in the ploidy reduction process of titan cells derived from a heterozygous diploid background in an important human fungal pathogen. SIGNIFICANCEPolyploid titan cells are central to Cryptococcus neoformans persistence during infection, yet how they reduce ploidy to generate daughter cells was unclear. In most eukaryotes, the meiosis-related proteins Spo11 and Dmc1 are best known for initiating (double-strand break formation) and executing (strand exchange) meiotic recombination, respectively. Here, we show that in C. neoformans, Dmc1 and Spo11 are critical for polyploid titan cells to produce viable daughter cells with reduced ploidy specifically in a heterozygous diploid context, ensuring genome stability without introducing detectable canonical meiotic recombination. Loss of either gene compromises daughter cell production and causes genome instability (LOH, aneuploidy) and phenotypic heterogeneity. These findings reveal a "para-meiosis" safeguard that preserves genomic configurations during titan cell depolyploidization in C. neoformans.

genetics↗

Single-cell morphodynamics predict cell fate decisions during mucociliary epithelial differentiation.

Cell state transitions underlie the emergence of diverse cell types and are traditionally defined by changes in gene expression. Yet these transitions also involve coordinated shifts in cell morphology and behaviour, which remain poorly characterized in densely packed epithelia. We developed a quantitative live-imaging and computational framework to track thousands of individual cells over time in the rapidly differentiating Xenopus mucociliary epithelium (MCE). From segmentations and trajectories, we extracted dynamic features-cell and nuclear shape, movement, position-to create a time-resolved morphodynamic dataset spanning the full course of differentiation. While single features showed high noise and low separability of ground-truth cell types, supervised machine learning revealed that integrating time-resolved features robustly predicts final cell fate. Gradient-boosted trees and multinomial logistic regression achieved moderate but consistent accuracy, especially for abundant epithelial lineages. Key discriminants included normalized Z-position, membrane-nucleus offset, and absolute experimental time, whereas movement contributed minimally to the results. Our results show that morphodynamic signatures encode predictive information about cell identity and provide a framework linking physical dynamics with molecular state.

developmental biology↗

BioLab: End-to-End Autonomous Life Sciences Research with Multi-Agents System Integrating Biological Foundation Models

Scientific discovery in the life sciences remains hindered by fragmented workflows, narrow-scope computational models, and inefficient links between in silico prediction and wet-lab validation. We present BioLab, a multi-agent system that integrates domain-specialized foundation models to automate end-to-end biological research. BioLab comprises eight collaborating agents, including a Planner, Reasoner, and Critic, orchestrated through a Memory Agent that enables iterative refinement via retrieval-augmented generation and a suite of 219 computational xBio-Tools spanning five biological scales (DNA, RNA, protein, cell, and chemical). These tools are built on the xTrimo Universe, a collection of 104 models derived from six foundation models (xTrimoChem, Protein, RNA, DNA, Cell, and Text), the majority of which achieve state-of-the-art (91-100% SOTA ratios) on domain benchmarks. Across standard reasoning tasks (PubMedQA, MMLU-Pro/Biology, GPQA-diamond), BioLab consistently outperformed leading large language models, including GPT-4o, Gemini-2.5, and DeepSeek-R1. Beyond benchmarks, BioLab autonomously executed a fully computational pipeline for de novo macrophage-targeting antibody design, progressing from target mining to multi-objective antibody optimization, where molecular dynamics simulations revealed structural mechanisms underlying enhanced affinity of optimized variants. Closing the computational-experimental loop, BioLab designed optimized antibodies (Pem-MOO-1, Pem-MOO-2) that achieved IC50 values of 0.01-0.016 nM, markedly surpassing the parental Pembrolizumab (0.027 nM) for PD-1. Functional assays confirmed enhanced pathway blockade and improved multi-parameter performance profiles. Together, these results establish BioLab as a generalizable framework for AI-native scientific discovery, demonstrating how multi-agent systems coupled with foundation models can autonomously generate, execute, and experimentally validate novel biological hypotheses.

bioinformatics↗

Core gastric microbiota linked to pathogenesis and preserved across age stratified cohorts

Mounting evidences reveal a complex microbial niche in gastric environment. Nevertheless, the phylogenetic stability of gastric microbiota remains undetermined. Here, using full-length 16S rRNA and ITS amplicon sequencing applied to gastric brushes from a pediatric endoscopic cohort (n=243), we identified a core microbiota comprising 11 genera, which clustered into two antagonistic communities (CST1A and CST1B). CST1A was enriched in healthy subjects, while CST1B correlated with gastroduodenal disease. Notably, this bipartite architecture arose through bacterial interactions, with fungal taxa demonstrating minimal integration. Further stratified analyses revealed that H. pylori infection preferentially depleted CST1A taxa. The subclusters exhibited divergent metabolic profiles, with CST1A specializing in motility processes and CST1B in nucleotide/peptidoglycan synthesis. Cross-cohort validation of six datasets (n=1,183) confirmed the persistent relevance of the CST1A-CST1B structure across disease stages. Collectively, our findings define a core gastric microbiota established in early life, linked to gastric disease risk, and persisting into adulthood. HighlightsO_LIIdentification of a gastric bacterial network established in early life. C_LIO_LIDysbiosis in two subclusters within the core microbiota is associated with pediatric peptic ulcers. C_LIO_LIHelicobacter pylori infection reduces the abundance of the core microbiota, with a particularly greater inhibitory effect on CST1A. C_LIO_LIThe core microbiota is preserved across age stratified cohorts and is associated with gastric cancer. C_LI

microbiology↗

CLM-access: A Specialized Foundation Model for High-dimensional Single-cell ATAC-seq analysis

Inspired by the success of large language models (LLMs) in natural language processing, cell language models (CLMs) have emerged as a promising paradigm for learning cell representations from high-dimensional single-cell data--particularly transcriptomic profiles from scRNA-seq. These foundation models have shown remarkable potential across a variety of downstream applications. However, there remains a lack of foundation models for scATAC-seq data, which measures chromatin accessibility at single-cell level and is critical for decoding epigenetic regulation. Developing such models is considerably more challenging due to the unique characteristics of scATAC-seq data, including the vast number of chromatin regions, lack of standardized annotations, extreme sparsity, and near-binary distributions. To address these challenges, we systematically explore various strategies and propose CLM-access, a specialized foundation model for scATAC-seq data. CLM-access incorporates three main innovations: (1) an unified data processing pipeline that maps 2.8 million cells onto an unified reference of over 1 million chromatin regions; (2) a specialized patching and embedding strategy to effectively manage high-dimensional inputs; and (3) a tailored masking and loss function design that preserves fine-grained regional information while enhancing training efficiency and representation quality. With comprehensive benchmarks, we show that CLM-access significantly outperforms existing methods in key downstream tasks, including batch effect correction, cell type annotation, RNA expression prediction, and multi-modal integration. This work establishes a scalable and interpretable foundation model for single-cell epigenomic analysis and expands the application of CLMs in single-cell research. Code is available at https://github.com/HIM-AIM/CLM-access

bioinformatics↗

BioinAI: a general bioinformatic framework for multi-level transcriptomic data analysis using multiple semi-agents

Clinical and biological insights from large-scale transcriptomic data are often limited by technical variability and analytical complexity. Here, we introduce BioinAI, a general bioinformatic framework designed for multi-level transcriptomic data analysis. As part of the BioinAI framework, two algorithms named DeepAdvancer and stNiche were developed to improve data integration and analytical efficiency. DeepAdvancer leverages a class-aware adversarial autoencoder to reconstruct gene-expression profiles from 49,738 samples across 131 skin conditions. These profiles revealed a conserved inflammatory axis and a transcriptomic continuum which link diverse diseases through shared immune responses and structural programs. stNiche leverages spatial graph networks and symmetry-aware matching to identify functional cellular niches and reveals microstructural alterations across development, homeostasis, and disease. For instance, it identified a fibroblast-immune niche surrounding hair follicles in healthy skin, which disappears in pathological states. In addition, BioinAI provides a user-friendly online analysis platform powered by multiple semi-agents (www.bioinai.com), significantly facilitating the extraction of biological insights to advance scientific research.

bioinformatics↗

Transient telomere uncapping triggers telomeric and subtelomeric rearrangements

Telomeres are nucleoprotein structures that cap the extremities of eukaryotic linear chromosomes, thus preventing them from being detected as DNA damage. Telomere uncapping poses a profound threat to genome integrity, yet the immediate consequences of transient uncapping remain unclear. In Saccharomyces cerevisiae, the Cdc13-Stn1-Ten1 complex caps telomeres and limits resection, which would otherwise lead to DNA damage checkpoint activation. Here, using the temperature-sensitive cdc13-1 allele, we demonstrate that even transient telomere uncapping induces extensive genomic rearrangements within a few cell cycles, despite a functional DNA damage checkpoint. Two distinct rearrangement signatures were observed in cells surviving transient uncapping: one characterized by the reorganization and recombination of the subtelomeric region, mostly involving the Y elements, and the other exhibiting massively elongated telomeres up to 10 kb, corresponding to a [~]30-fold increase. Long-read sequencing revealed that the genomic instability was confined to the subtelomere and telomere regions, and evidenced Y' element loss/amplification, terminal duplication of chromosome ends, and telomeric-circle-driven amplification of telomere repeats. Rearrangements unfold over multiple generations and require the homologous recombination factor Rad52 and the Pol{delta} subunit Pol32, which is essential for break-induced replication. The recombination proteins Rad51 and Rad59 also contribute to the rearrangements in partially independent pathways. Remarkably, survivors with elongated telomeres demonstrate robust resistance to subsequent telomere uncapping, in a Rad52-dependent manner. Our findings provide novel insights into the consequences of transient telomere uncapping for genome stability, a process that might contribute to subtelomere and telomere dynamics and evolution.

molecular biology↗

Genomic Surveillance Reveals Global Spread of Macrolide-Resistant Bordetella pertussis Linked to Vaccine Changes

The resurgence of whooping cough in regions utilizing acellular pertussis vaccines underscopes emerging public health challenges. Here, we characterized 178 Bordetella pertussis isolates collected from patients across all age groups in Shanghai (2018-2024) to assess genomic evolution and antibiotic susceptibility. Macrolide resistance to erythromycin, azithromycin, clarithromycin and clindamycin escalated from [≤]50% (pre-2020) to nearly 100% (post-2020), mechanistically linked to the 23S rRNA A2047G mutation. Genome-based analysis identified a genotype MT28-ptxP3-MRBP rapidlly dominated post-2020, exhibiting significantly higher prevalence in adults versus than age groups. Phylogenetic analysis of 178 Shanghai and 1596 global genomes revealed two major lineages corresponding to ptxP1 and ptxP3 alleles. MT28-ptxP3-MRBP cluster was identified in France, Japan and the United States in 2024, indicating potential cross-border dissemination. These findings advocate for intergrated surveillance spanning all ages and international borders to contain the global spread of macrolide-resistant Bordetella pertussis. HighlightsO_LIAfter 2020, MT28-ptxP3-MRBP lineage rapidly dominated, comprising 61.7% of isolates. C_LIO_LIMT28-ptxP3-MRBP exhibits a significant transmission advantage among older individuals. C_LIO_LIThe primary affected group shifted from [≤]36 months (pre-2020) to 37 months-18 years (post-2020). C_LIO_LIMacrolide resistance rose from [≤]50% pre-2020 to nearly 100% post-2020, with all resistant isolates carrying the A2047G mutation. C_LI

microbiology↗

EVscope: A Comprehensive Bioinformatics Pipeline for Accurate and Robust Analysis of Total RNA Sequencing from Extracellular Vesicles

MotivationExtracellular vesicle (EV) RNA sequencing has emerged as a powerful approach for studying RNA biomarkers and intercellular communication. Nevertheless, the extremely low abundance, fragmented nature and ubiquitous tissue origin of EV RNAs, alongside potential contamination from co-isolated materials, such as free DNA and bacterial RNA, pose substantial analytical challenges. These complexities highlight a pressing need for a standardized, computational workflow that ensures robust quality control and EV RNA characterization. ResultsHere, we present EVscope, an open-source bioinformatics pipeline designed specifically for processing EV RNA-seq datasets. EVscope employs an optimized genome-wide expectation-maximization (EM) algorithm that significantly improves multi-mapping read assignment at single-base resolution by effectively leveraging alignment scores (AS) and local read coverage, specifically tailored for fragmented and low-abundance EV RNAs. Notably, EVscope uniquely generates EM-based BigWig files for downstream analysis, a capability currently unavailable in existing EM-based BigWig quantification tools. The pipeline systematically integrates 27 major steps, including quality control, analysis of library structure, contamination assessment, read alignment, read strandedness detection, UMI-based deduplication, RNA quantification, genomic DNA (gDNA) contamination correction, cellular and tissue source inference and visualization with a comprehensive HTML report. EVscope incorporates a comprehensive, updated annotation covering 19 distinct RNA biotypes, encompassing protein-coding genes, lncRNAs, miRNAs, piRNAs, retrotransposons (LINEs, SINEs, ERVs), and additional non-coding RNAs (tRNAs, rRNAs, snoRNAs). Furthermore, it leverages two highly balanced circRNA detection algorithms for robust circular RNA identification. Notably, a downstream module enables the inference of the tissue/cellular origins of EV RNAs using bulk and single-cell RNA-seq reference datasets. EVscope is implemented as a convenient, single-command Bash pipeline leveraging Conda-managed standard software packages and custom scripts, ensuring reproducibility and straightforward deployment. Availability and implementationCode, documentation, and tutorials are available at GitHub (https://github.com/TheDongLab/EVscope) and archived on Zenodo (https://zenodo.org/records/15577789).

bioinformatics↗

Simulation and experimental study on the strategy of soil auxiliary heating in a solar greenhouse in northeast China

When solar greenhouses in north eastern China experience extreme weather, the internal soil requires varying auxiliary heating to ensure the normal growth of crops. In this study, considering the effects of four kinds of extreme weather on solar greenhouses, four different types of heating requirements were obtained through experiences in a Liaoshen-type solar greenhouse. Four distinct heating strategies were simulated using a soil temperature prediction model. The initial temperatures for these strategies were 13{degrees}C, 11{degrees}C, 13{degrees}C, and 14{degrees}C, respectively, while the target temperature was uniformly set at 15{degrees}C. The heating durations for the respective strategies were 432,000 seconds, 432,000 seconds, 864,000 seconds, and 1,728,000 seconds. To ensure a minimum soil temperature of 15{degrees}C within the solar greenhouse, the temperature of the heating layer located 400 mm underground needed to be maintained at or above 20{degrees}C, 28{degrees}C, 18{degrees}C, and 19{degrees}C, respectively. The simulation results were validated through experimental verification, with maximum and minimum errors of 3.1% and 0.5%, respectively. Several targeted promotional suggestions were also proposed. The model exhibited high accuracy and is capable of developing optimal and precise soil heating strategies for various types of solar greenhouses in north eastern China.

plant biology↗

Metabolite Interactions Mediate Beneficial Alliances Between Bacillus and Trichoderma for Effective Fusarium Wilt Control

Bacteria-Fungi Interactions play a crucial role in soil nutrient cycling and plant disease suppression. Bacillus and Trichoderma exhibit antagonism when inoculated on laboratory media, global soil sample analysis reveals a positive correlation between these two genera in addition to enhanced plant-pathogen Fusarium oxysporum (FOC) suppression and plant growth promotion. Here, we depict a complexity of interactions in a cross-kingdom consortia of Bacillus velezensis and Trichoderma guizhouense. Transcriptomic profiling revealed that in the presence of fungi, the key stress sigma factor of B. velezensis activates expression of biosynthetic genes for antimicrobial secondary metabolite production. Among these, surfactin induces T22azaphilone production in T. guizhouense that hinders oxidative stress. Both surfactin and T22azaphilone contribute to Bacillus and Trichoderma maintenance in soil in the presence of FOC. Finally, FOC-secreted fusaric acid temporarily inhibits B. velezensis growth while it is efficiently degraded by T. guizhouense. These metabolite-mediated interactions reveal how competing soil microorganisms could form effective alliances that ultimately enhance plant protection against soil-borne pathogens.

microbiology↗

Predicting the Evolutionary and Functional Landscapes of Viruses with a Unified Nucleotide-Protein Language Model: LucaVirus

Predicting viral evolution and function remains a central challenge in biology, hindered by high sequence divergence and limited knowledge compared to cellular organisms. Here, we introduce LucaVirus, a multi-modal foundation model for viruses, trained on 25.4 billion nucleotide and amino acid tokens covering nearly all known viruses. LucaVirus learns biologically meaningful representations capturing relationships between sequences, protein/gene homology, and evolutionary divergence. Using these embeddings, we developed downstream models that address key virology tasks: identifying hidden viruses in genomic "dark matter", annotating enzymatic activities of uncharacterized proteins, predicting viral evolvability, and identifying antibody candidates for emerging viruses. LucaVirus achieves state-of-the-art results in three tasks and matches leading models in the fourth with one-third the parameters. Together, these findings demonstrate the power of a unified foundation model to comprehensively decode the viral world and establish LucaVirus as an efficient and versatile platform for AI-driven virology, from virus discovery to functional and therapeutic predictions.

bioinformatics↗