Biology subjects

Zhang, Z.

Publications and source records attributed to Zhang, Z..

At least 361 records · Page 20Linked to original sources

bioRxiv2025

BCKDK, a novel hypoxia-responsive kinase that exacerbates cerebral ischemia injury

Alterations of circulating amino acid profile have been observed in patients with ischemic stroke. However, whether ischemia disrupts amino acid metabolism in the brain tissue and subsequently potentiates cellular stress and cerebral injury have never been explored. Employing a metabolomics approach combined with metabolic flux analysis, impaired catabolism and significant enrichment of branched-chain amino acids (BCAAs) were identified in mouse primary neuron cells upon oxygen-glucose deprivation. Consistently, BCAA catabolism was also damaged in the brain of mouse with acute ischemic stroke, accompanied with suppressed activity of branched-chain alpha-keto acid dehydrogenase (BCKDH) and upregulation of BCKDH kinase (BCKDK). Furthermore, restoration of BCAA catabolism by suppressing BCKDK via pharmacological inhibitor or silencing RNA dramatically alleviated cerebral ischemia injury in mice. Mechanistically, ischemia induces the expression of BCKDK via hypoxia-inducible factor 1-mediated transcriptional activation and inhibits BCAA conversion into substrates for tricarboxylic acid cycle, contributing to potentiated energy deficiency, glutamate excitotoxicity, and neuronal injury. Collectively, this study identified BCKDK as a novel hypoxia-responsive factor and a promising therapeutic target for cerebral ischemia injury.

molecular biology↗

bioRxiv2025

Divergent spontaneous antibiotic-resistance evolution confers reciprocal and exploitable collateral sensitivity effects

The global rise of antibiotic-resistant pathogens has outpaced the development of new antibiotics, prompting the urgent need for alternative treatment strategies. One such approach is to leverage collateral sensitivity (CS), where resistance to one antibiotic increases susceptibility to another. However, the clinical implementation of CS-based therapies depends on the consistency of these responses, which is challenged by variable resistance mutations and the dynamics of resistant strains during infection. Here, we combined experiments and mathematical models to assess the consistency and consequences of CS responses in the Gram-positive pathogen Streptococcus pneumoniae following the de novo acquisition of resistance to five commonly used antibiotics. We found that many collateral responses were unpredictable and inconsistent between different resistance mutations. However, for two antibiotic pairs, we identified consistent unidirectional (RIF [->] FUS) and bidirectional (LNZ {leftrightarrow} FUS) CS interactions, despite the divergent evolutionary trajectories of resistant strains, as revealed by whole-genome sequencing. To evaluate if CS for these combinations can be exploited to design dosing strategies to eradicate S. pneumoniae infections while suppressing resistance, we developed a mathematical stochastic pharmacokinetic-pharmacodynamic (PK-PD) model, which integrated our experimentally derived PD parameters with existing clinical PK models. Our model-based analyses confirmed the superiority of these antibiotic combinations over monotherapy and showed that their efficacy depends on the presence of CS interactions between the administered antibiotics. In summary, our study demonstrates how consistent CS interactions can be leveraged to inform treatment strategies, laying the groundwork for CS-guided therapies to preserve antibiotic efficacy.

microbiology↗

bioRxiv2025

Discovery of circulating cell-free DNA 5hmC biomarkers for peritoneal metastasis in colorectal and appendiceal cancer

IntroductionPeritoneal metastases (PM) are associated with poor prognosis in patients with colorectal cancer (CRC) or appendiceal adenocarcinoma (AA), yet detection of PM is unreliable using current circulating DNA technology. Leveraging novel 5hmC-seal technology to detect ultra-low amounts of DNA in plasma, we demonstrate the feasibility of 5-hydroxymethylcytosine (5hmC) signatures derived from circulating cell-free DNA (cfDNA) as biomarkers for PM. MethodsUsing a highly sensitive and robust 5hmC sequencing approach on genomic DNA isolated from peripheral blood samples, we developed predictive models to identify biomarkers for peritoneal metastases. ResultsWe obtained genome-wide 5hmC profiles from 71 CRC/AA patients with PM, 41 without PM, and 73 non-cancer controls. Predictive models trained on genomic region 5hmC levels in patients with cancer could distinguish PM status with high sensitivity and moderate specificity (AUC 0.827, sensitivity 92.4%, specificity 46.1%). Pathway enrichment analysis identified epigenetically dysregulated cancer, cell migration, adhesion, and immune-related pathways in PM. ConclusionNovel 5hmC-Seal technology based 5hmC signatures can detect patients with peritoneal metastases from CRC and AA, albeit with reduced specificity. This study lays a foundation for future clinical assay development for PM. Statement of significanceWe demonstrate high-sensitivity detection of peritoneal metastasis in colorectal and appendiceal adenocarcinomas using 5hmC-Seal of plasma cfDNA. Earlier detection of this condition could expand curative treatments in [~]20,000 affected U.S. patients.

cancer biology↗

bioRxiv2025

3D-printing-assisted, microfabricated devices reveal hierarchical and temporal mechanosensing in high-density fibroblast culture

Understanding how cells integrate mechanical forces across multiple directions, length scales, and timescales remains a fundamental challenge in mechanobiology. Deciphering how cells integrate this information is particularly important in the context of wound healing, where the timing and duration of the fibroblast-to-myofibroblast transition can determine healing outcomes. Here, we discovered that fibroblasts in engineered tissues respond to directional anisotropy in stress through a hierarchical temporal cascade, with individual cell elongation (24 hr) preceding collective alignment (40 hr), which then drives -smooth muscle actin expression and myofibroblast transition (96h). To enable this discovery, we developed a modified hydrogel-assisted stereolithographic elastomer (HASTE) prototyping platform to incorporate a detergent that improves wettability of template agar hydrogels by poly(dimethylsiloxane) elastomer. HASTE allowed rapid prototyping of intricate 3D micropost arrays that provides isotropic (8-post) versus anisotropic (4-post) boundary conditions. Fibroblasts sensed and responded to stress directionality before bulk tissue reorganization occurs. Computational modeling predicted steady-state activation patterns based on initial stress anisotropy rather than magnitude, and our experiments reveal that reaching this state requires sequential mechanosensitive processes operating across distinct timescales. This temporal hierarchy persists even when extensive cell-cell contacts might be expected to mask matrix-mediated mechanical signals. Our findings demonstrate that fibroblast mechanosensing involves adaptive responses encoded through progressive cell and tissue reorganization. Results provide insight into how nanoscale mechanosensing scales up to direct tissue-level organization, with implications for understanding wound healing, understanding fibrosis, and engineering functional tissue replacements.

bioengineering↗

bioRxiv2025

Endothelial CEPT1 Regulates Hepatic MTTP-Mediated Lipid Metabolism and Impacts Aortic Atherosclerosis

BackgroundThe regulation of hepatic lipid metabolism by vascular endothelial factors remains poorly characterized, despite its relevance to atherosclerosis and steatosis. Microsomal triglyceride transfer protein (MTTP) is essential for hepatic lipid metabolism, but its regulation by endothelial cells has not been previously investigated. ObjectiveThis study examined whether endothelial choline ethanolamine phosphotransferase 1 (CEPT1) modulates hepatic MTTP activity, impacting systemic lipid homeostasis and aortic plaque formation. Methods and ResultsHuman steatotic liver samples exhibited reduced CEPT1 and MTTP protein levels, correlating with diminished lipid exports. In mice, endothelial-specific Cept1 knockdown decreased hepatic MTTP expression, reduced serum triglyceride and cholesterol levels, and markedly attenuated aortic atherosclerosis without evidence of fat malabsorption. In vitro, endothelial CEPT1 silencing suppressed MTTP activity in co-cultured hepatocytes via a paracrine mechanism involving peroxisome proliferator-activated receptor (PPAR) signaling, which was rescued by fenofibrate treatment. Aortic histology confirmed reduced plaque burden and macrophage infiltration in CEPT1-deficient mice. ConclusionsEndothelial CEPT1 critically regulates hepatic MTTP through a paracrine axis, influencing lipid metabolism and atherogenesis. Targeting endothelial CEPT1 may represent a novel therapeutic approach to reduce steatosis and vascular atherosclerosis.

molecular biology↗

bioRxiv2025

Label-Free, Real-Time, In Vivo Optical Biopsy with a Handheld Quantitative Phase Microscope

In this work we develop and demonstrate the utility of a compact, handheld quantitative phase imaging microscope that enables label-free, in vivo optical imaging of bulk tissues with clear cellular and subcellular histological detail in real-time. The proposed device overcomes significant challenges in optical imaging for in vivo applications, particularly for clinical human use. The approach uses quantitative oblique back illumination microscopy (qOBM) to obtain quantitative phase information of opaque samples using epi-illumination. The compact handheld probe achieves 0.8 {micro}m lateral resolution, 5 {micro}m axial resolution, 300 {micro}m X 300 {micro}m field of view, and operates at 25Hz in a wide-field (non-scanning) configuration, enabling real-time imaging. The probe is also inexpensive and has no moving components, making it robust. The utility of the probe is demonstrated in (1) human skin in vivo, (2) brain tumor tissue ex vivo from a murine tumor model and from discarded human tissue from neurosurgery, and (3) in vivo using healthy brain tissue from a large animal model (swine), simulating neurosurgical conditions. Given the clear cellular and subcellular histological detail (i.e., "optical biopsy") obtained in real-time, combined with the ease-of-use and low-cost of the system, the proposed device has significant implications for a broad range of clinical applications.

bioengineering↗

bioRxiv2025

A BIOSECURITY AGENT FOR LIFECYCLE LLM BIOSECURITY ALIGNMENT

AO_SCPLOWBSTRACTC_SCPLOWLarge language models (LLMs) are increasingly integrated into biomedical re-search workflows-from literature triage and hypothesis generation to experimental design-yet this expanded utility also heightens dual-use concerns, including the potential misuse for guiding toxic compound synthesis. In response, this study shows a Biosecurity Agent that comprises four coordinated modes across the model lifecycle: dataset sanitization, preference alignment, runtime guardrails, and automated red teaming. For dataset sanitization (Mode 1), evaluation is conducted on CORD-19, a COVID-19 Open Research Dataset of coronavirus-related scholarly articles. We define three sanitization tiers-L1 (compact, high-precision), L2 (human-curated biosafety terms), and L3 (comprehensive union)-with removal rates rising from 0.46% to 70.40%, illustrating the safety-utility trade-off. For preference alignment (Mode 2), DPO with LoRA adapters internalizes refusals and safe completions, reducing end-to-end attack success rate (ASR) from 59.7% to 3.0%. At inference (Mode 3), runtime guardrails across L1-L3 show the expected security-usability trade-off: L2 achieves the best balance (F1 = 0.720, precision = 0.900, recall = 0.600, FPR = 0.067), while L3 offers stronger jailbreak resistance at the cost of higher false positives. Under continuous automated red-teaming (Mode 4), no successful jailbreaks are observed under the tested protocol. Taken together, our biosecurity agent offers an auditable, lifecycle-aligned framework that reduces attack success while preserving benign utility, providing safeguards for the use of LLMs in scientific research and setting a precedent for future agent-level security protections.

bioinformatics↗

bioRxiv2025

Dynamic Regulation of the Immune Repertoire of Bacteria

The CRISPR-Cas system provides adaptive immunity in many bacteria and archaea by storing short fragments of viral DNA, known as spacers, in dedicated genomic arrays. A longstanding question in CRISPR-virus coevolution is the optimal number of spacers for each bacterium to maintain proper phage coverage. In this study, we investigate the optimal CRISPR memory size by combining steady-state immune models with dynamical antigenic traveling wave theory to obtain both analytic and numerical results of coevolutionary dynamics. We focus on two experimentally supported phenomena that shape immune dynamics: primed acquisition, where partial spacer-protospacer matches boost acquisition rates, and cassette expansion, where a short-term increase in memory size drives population dynamics. We find that under primed acquisition, longer optimal arrays benefit from maintaining multiple, partially matching spacers. In contrast, dynamic cassette expansion favors shorter arrays by amplifying the fitness advantage of acquiring a few highly effective new spacers. Together, our results highlight that memory optimality is not fixed, but instead shaped by the interaction of acquisition dynamics and population-level immune pressures.

biophysics↗

bioRxiv2025

Host Phosphatase PPM1G Governs Influenza vRNP Homeostasis by Orchestrating NP Phosphorylation Dynamics and Autophagy

The influenza viral ribonucleoprotein (vRNP) serves as the machinery for viral RNA replication. It self - assembles from the RNA - dependent RNA polymerase (RdRp), viral RNA, and nucleoprotein (NP). During vRNP replication, NP proteins must rapidly switch between monomeric and polymeric states to sustain dynamic equilibrium, which relies on NP phosphorylation-dephosphorylation cycling and is crucial for viral nucleic acid replication. However, the regulatory mechanisms controlling this process remain poorly understood. In this study, we employed an RdRp/vRNP-targeted affinity mass spectrometry-based differential analysis method, discovered that the cellular protein phosphatase Mg2+/Mn2+-dependent 1G (PPM1G) is a phosphatase of NP and required for maintaining viral polymerase activity. Overexpression of PPM1G decreased viral replication. In ppm1g flox/flox-Sftpc Cre mice, lethal influenza infection led to survival and mild disease, as well as reduced lung viral loads and inflammation. These results suggest that PPM1G is a key host molecule for maintaining steady state of viral replication. PPM1G expression is upregulated upon influenza virus infection. This heightened expression occurs in response to increased viral protein production, simultaneously leading to elevated dephosphorylation of NP, which accelerates NP polymerization to promote viral replication. Meanwhile, surplus NP can be degraded through the ATG7 autophagylysosome pathway. Our research elucidates a mechanism that PPM1G maintains viral replication homeostasis by regulating NP status and fate.

microbiology↗

bioRxiv2025

Single-nucleus transcriptomics reveal disrupted pathways in the prefrontal cortex of Scn2a-deficient mice

Truncating variants in SCN2A, which encodes the NaV1.2 sodium channel critical for action potential initiation and propagation, are associated with autism spectrum disorder (ASD) and epilepsy. To investigate SCN2A deficiency-related phenotypes, we developed a preclinical mouse model with under 50% NaV1.2 expression, exhibiting neuronal hyperexcitability and social deficits. However, the neuronal populations and molecular alterations underlying these phenotypes at single-cell resolution have not been investigated. In this study, we conducted single-nucleus RNA sequencing (snRNA-seq) of wild-type (WT), homozygous Scn2a-deficient (HOM) mice, and HOM mice with Scn2a restoration (HOM-FlpO) to examine the effects of Scn2a level on gene expression in the medial prefrontal cortex (mPFC), a critical brain region related to ASD development. Differential expression analysis in GABAergic and glutamatergic neurons between genotypes revealed gene expression enriched in neurotransmitter activity regulation and synapse organization. Lastly, snRNA-seq results in HOM-FlpO identified genes that were rescued after Scn2a restoration. These results reveal that reduced Scn2a expression disrupts RNA transcriptomes in multiple neuronal subtypes, providing insight into cell type-specific mechanisms underlying SCN2A-related disorders.

neuroscience↗

bioRxiv2025

Metabolic Signatures of Pulmonary Embolism in COVID-19: Insights from Longitudinal Intensive Care Unit Profiles

ABSTRACTO_ST_ABSBackgroundC_ST_ABSPulmonary embolism is a severe complication of COVID-19 infection, associated with a hypercoagulable state and heightened risk of blood clots. As SARS-CoV-2 has become endemic, understanding pulmonary embolisms metabolic effects in COVID-19 patients is warranted. This study investigated the longitudinal metabolic profiles of 66 Intensive Care Unit-admitted COVID-19 patients at Erasmus Medical Center to identify metabolites and mechanisms associated with pulmonary embolism. MethodA total of 1209 metabolic species were measured, including amines and lipids. Metabolic changes were analysed across four timeframes: i) general analysis of pulmonary embolism, ii) 72 hours prior to pulmonary embolism, iii) 48 hours prior and the day of pulmonary embolism, and iv) the day of and 48 hours post-pulmonary embolism. ResultsThe general analysis revealed significant upregulation of amines, triglycerides, phosphatidylethanolamines, ether-linked phosphatidylethanolamines, and eicosanoids in patients who developed a pulmonary embolism. Phosphatidylethanolamines containing the 20:3 fatty acid side chain were notably elevated. Minimal metabolic dysregulation was observed 72 hours before pulmonary embolism, with subtle increases in lysophosphatidylcholines and lysophosphatidylethanolamines. In contrast, there was a strong metabolic response during and post-pulmonary embolism, phosphatidylethanolamines (47%), ether-linked phosphatidylethanolamines(96%) and sphingosines(40%). ConclusionThese findings underscore the critical role of lipid metabolism in pulmonary embolism, particularly triglycerides and specific lipid species. The limited metabolic perturbations before pulmonary embolism suggest early prediction challenges, emphasising the need for further research into temporal metabolic changes and their clinical applications.

molecular biology↗

bioRxiv2025

Mitotic CDK4/6 activity sustains spindle checkpoint signalling to prevent mitotic slippage and genomic instability

The precise regulation of cell cycle entry and the maintenance of genome integrity are crucial for preventing tumorigenesis. Cyclin-dependent kinases 4 and 6 (CDK4/6) play pivotal roles in linking mitogenic signals to G1-S phase progression1-3, and their frequent deregulation in various cancers underscores their importance in driving cell proliferation and as therapeutic targets4,5. Despite this, the roles of CDK4/6 beyond the G1/S transition remain underexplored. Here, we uncover a previously unrecognised function of CDK4/6 in mitotic progression through regulation of the spindle assembly checkpoint (SAC)6. Using both cancer and non-transformed human cells, we show that acute CDK4/6 inhibition after G1/S transition leads to premature mitotic exit despite unattached kinetochores, resulting in chromosome missegregation and aneuploidy. Phosphoproteomic analyses and in vitro kinase assays reveal that CDK4 phosphorylates multiple sites on key SAC regulators, including the C-terminal tail of CENP-E, which is critical for BubR1 recruitment to kinetochores and SAC maintenance7,8. CDK4/6 inhibition reduces the phosphorylation of SAC components, attenuating checkpoint signalling and accelerating mitotic slippage. Notably, residual SAC activity persists, suggesting that CDK4/6 strengthens or stabilises SAC signalling rather than being essential for its activation. Thus, CDK4/6 functions extend beyond G1/S control to ensure mitotic fidelity, linking proliferating signals to genome stability and exposing potential vulnerabilities to anti-mitotic therapies in cancer.

cell biology↗

bioRxiv2025

The two-component nuclease-active KELShedu system confers broad antiphage activity via abortive infection

Bacteriophages and bacteria engage in a continuous evolutionary arms race, driving the development of intricate bacterial defense systems such as CRISPR-Cas, BREX, Gabija, and Shedu. Here, we characterize a two-component KELShedu system in Escherichia coli that confers resistance to phages via abortive infection. The KELShedu system comprises KELA, a dsDNA-binding protein, and KELB, a metal ion-dependent nuclease harboring the DUF4263 domain. In addition, we find that physiological levels of NTP inhibit the DNA cleavage activity of the KELShedu system, suggesting that KELShedus activation depends on reduced intracellular NTP levels during phage invasion. Our research demonstrates that the KELShedu system responds to nucleotide depletion triggered by phage replication, leading to non-specific degradation of cellular DNA and ultimately inducing abortive infection. These insights into the KELShedu system expand the repertoire of bacterial anti-phage mechanisms and lay the groundwork for novel applications in microbial engineering and therapeutic development. TeaserThe KELShedu system exerts anti-phage function by aborting infection in response to physiological changes following phage infection

microbiology↗

bioRxiv2025

qcMol: a large-scale dataset of 1.2 million molecules with high-quality quantum chemical annotations for molecular representation learning

AO_SCPLOWBSTRACTC_SCPLOWRecent advancements in deep learning have greatly prompted the de novo design of drugs and materials. Previous studies have shown that a well-designed molecular representation is critical for improving the accuracy of deep-learning-based molecular property prediction methods. However, the lack of large-scale data enriched with detailed physicochemical information hinders effective learning of an informative molecular representation. To fill this data gap, we introduce qcMol, a dataset consisting of 1.2 million molecules from 95 datasets with high-quality quantum chemical annotations, to facilitate molecular representation learning as well as downstream molecular property prediction. Chemicals in this dataset include drug-like compounds, metabolites and molecules with matched experimental data, covering 247,448 kinds of scaffolds and a broad spectrum of molecular sizes. Each compound in qcMol is annotated with detailed quantum chemical information, obtained through reliable quantum chemical calculations based on B3LYP-D3/def2-SV(P)//GFN2-xTB as well as the follow-up wave function post-analysis. These features are organized into multiple formats, allowing for flexible integration into diversified molecular representation learning frameworks. The broad data distribution, comprehensive quantum chemical annotations and flexible data formats jointly enable qcMol to serve as the pre-training resource as well as the benchmark test set for deep learning models, benefiting the practical in silico drug discovery. qcMol is freely accessible from https://structpred.life.tsinghua.edu.cn/qcmol/.

biophysics↗

bioRxiv2025

Riemannian Diffusion Kernel-smoothed Continuous Structural Connectivity On Cortical Surface

Atlas-free continuous structural connectivity has garnered increasing attention due to the limitations of atlas-based approaches, including the arbitrary selection of brain atlases and potential information loss. Typically, continuous structural connectivity is represented by a probability density function, with kernel density estimation as a common estimation method. However, constructing an appropriate kernel function on the cortical surface poses significant challenges. Current methods often inflate the cortical surface into a sphere and apply the spherical heat kernel, introducing distortions to density estimation. In this study, we propose a novel approach using the Riemannian diffusion kernel derived from the Laplace-Beltrami operator on the cortical surface to smooth streamline endpoints into a continuous density. Our method inherently accounts for the complex geometry of the cortical surface and exhibits computational efficiency, even with dense tractography datasets. Additionally, we investigate the number of streamlines or fiber tracts required to achieve a reliable continuous representation of structural connectivity. Through simulations and analyses of data from the Adolescent Brain Cognitive Development (ABCD) Study, we demonstrate the potential of the Riemannian diffusion kernel in enhancing the estimation and analysis of continuous structural connectivity.

neuroscience↗

bioRxiv2025

A simplified and highly efficient cell-free protein synthesis system for prokaryotes

Cell-free protein synthesis (CFPS) systems are a powerful platform with immense potential in fundamental research, biotechnology, and synthetic biology. Conventional prokaryotic CFPS systems, particularly those derived from Escherichia coli (E. coli), often rely on complex reaction buffers containing up to thirty-five components, limiting their widespread adoption and systematic optimization. Here, we present an optimized E. coli cell-free protein synthesis (eCFPS) system, which is significantly streamlined for high efficiency. Through systematic screening, we successfully reduced the essential core reaction components from 35 to a core set of 7. The thorough optimization of these seven key components ensured that protein expression levels were not only maintained but even substantially improved. Furthermore, we developed a much simpler procedure for preparing the bacterial cytosolic extracts, a "fast lysate" protocol that eliminates the traditional time-consuming runoff and dialysis steps, thereby enhancing the overall accessibility and robustness of eCFPS. This optimized and user-friendly eCFPS efficiently synthesizes challenging proteins, including functional, self-assembling vimentin, and active restriction endonuclease BsaI despite its strong cytotoxicity, and serves as a powerful tool that will facilitate diverse applications in basic life science research and beyond.

biochemistry↗

bioRxiv2025

Scaling Personalized TMS: A Scalp-Based MRI-guided Alternative to Neuronavigation

BackgroundPersonalized transcranial magnetic stimulation (TMS) has shown promise in treating psychiatric and neurological disorders, but its accessibility is limited by reliance on costly and complex neuronavigation systems, particularly in resource-constrained clinical settings. ObjectiveTo develop and validate a scalp-based alternative to neuronavigation for personalized TMS targeting. MethodsWe developed a Continuous Proportional Coordinate (CPC) system-based approach for personalized TMS targeting and validated its performance through three operators across ten participants, using individualized DLPFC-SGC targets as an example. The validation framework included targeting consistency assessment, reproducibility evaluation, and cross-cortical generalizability analysis. ResultsThe CPC-based approach demonstrated comparable performance to neuronavigation, achieving 99.4% of its electric field strength (74.80 vs. 75.14 V/m; r = 0.962) and 93.9% of DLPFC-SGC functional connectivity (-0.216 vs. -0.230; r = 0.949), with high inter-operator reliability (ICC [≥] 0.903). Reproducibility analysis showed strong correlation between sessions (r = 0.948-0.957) with high reliability. Further simulations across 1,125 uniformly distributed cortical targets supported generalizability of this approach, with stable electric field preservation (99.07 {+/-} 1.54%) across different brain regions. ConclusionsThe CPC-based approach offers a practical alternative to neuronavigation, maintaining targeting accuracy and providing a more accessible option for personalized TMS implementation.

neuroscience↗

bioRxiv2025

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↗