LabOS: The AI Co-Scientist That Sees and Works With Humans

Cong, L.; Zhang, Z.; Wang, X.; Di, Y.; Jin, R.; Gerasimiuk, M.; Wang, Y.; Dinesh, R. K.; Smerkous, D.; Smerkous, A.; Wu, X.; Liu, S.; Li, P.; Zhu, Y.; Serrao, S.; Zhao, N.; Mohammad, I. A.; Sunwoo, J. B.; Wu, J. C.; Wang, M.

doi:10.1101/2025.10.16.679418

Home/Archive/bioRxiv

bioRxiv · 10.1101/2025.10.16.679418

LabOS: The AI Co-Scientist That Sees and Works With Humans

View original source ↗Export metadata ↓

Abstract

Modern science advances fastest when thought meets action. LabOS represents the first AI co-scientist that unites computational reasoning with physical experimentation through multimodal perception, self-evolving agents, and XR-enabled, embodied human-AI collaboration. To empower agentic AI with embodied intelligence, we built LabSuperVision, the first vision-language-model (VLM) benchmark spanning biomedical and materials science laboratories, designed to evaluate scientific perception and reasoning. This endowed LabOS with visual reasoning capabilities that surpass those of general-purpose models on laboratory-specific tasks. By connecting multi-model AI agents, smart glasses, and human-AI-robot collaboration, LabOS allows AI to see what scientists see, understand experimental context, and participate in real-time execution through immersive XR interaction and robotic autonomation. Across applications-from cancer immunotherapy target discovery to cell engineering-LabOS shows that AI can move beyond computational design to participation, turning the laboratory into an intelligent, collaborative environment where human and machine discovery evolve together.

Explore related subjects

bioengineering Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Cong, L., Zhang, Z., Wang, X., Di, Y., Jin, R., Gerasimiuk, M., Wang, Y., Dinesh, R. K., Smerkous, D., Smerkous, A., Wu, X., Liu, S., Li, P., Zhu, Y., Serrao, S., Zhao, N., Mohammad, I. A., Sunwoo, J. B., Wu, J. C., Wang, M.. 2025-10-19. LabOS: The AI Co-Scientist That Sees and Works With Humans. https://doi.org/10.1101/2025.10.16.679418

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Comparative study of chlorophyll measurement in Physcomitrium patens moss using a conventional microscope adapted for combined 2D+1D imaging and spectral analysis

Imaging spectroscopy often requires expensive and complex equipment. Here we show a simple procedure for attaching a standard miniature fiber spectrometer to a conventional microscope, allowing easy integration of 2D imaging with 1D high-resolution spectral measurements. This combination provides much of the benefit of a full imaging spectrometer without the large equipment investment, and we provide instructions for modifying microscopes to this setup and the present measurements of living cells that demonstrate their performance. Using this setup, we compare the quantitative measurement of chlorophyll concentration in Physcomitrium patens moss using color imaging and spectral sampling.

bioengineering↗

bioRxiv2026

De novo designed single-domain antibodies protect against lethal cobra venom neurotoxicity in vivo

Generative protein design can now rapidly produce de novo binders with high affinity and functional activity against a wide range of targets, including lethal snake venom toxins. However, so far most reported successes rely on new-to-nature scaffolds with limited therapeutic precedent. Single-domain antibodies (VHHs) offer a clinically validated alternative scaffold that can bind and neutralize long-chain -neurotoxins, which are some of the most lethal components in snake venoms. Here we compare three recently established de novo design models with VHH-design capabilities (Germinal, RFantibody, and BoltzGen) for their ability to generate VHHs against the neurotoxin -cobratoxin from the monocled cobra (Naja kaouthia). Using standardized model inputs and evaluation criteria based on AlphaFold3 interface confidence (ipTM) and RMSD self-consistency, we find that Germinal was the only method to generate designs passing stringent in silico criteria for experimental testing. We therefore performed a larger Germinal design campaign employing three different VHH frameworks and experimentally validated 46 designs in vitro. Of these, 42 expressed as soluble proteins and we identified four binding hits derived from two of the three tested frameworks. Of the four binders, two lead candidates were further characterized and demonstrated high affinity (KDs of 4.1 nM and 10.8 nM), monomeric behavior and low polyreactivity, indicating favorable biophysical and developability properties, as well as functional toxin neutralization in vitro. To assess their therapeutic potential we investigated their ability to protect against -cobratoxin toxicity in vivo. Both candidates fully protected mice after -cobratoxin challenge, with 100% survival compared to a lethal control. One candidate also retained notable neutralization capacity against whole venom of Naja kaouthia with a survival of 56%, while the other protected 22% when tested in a rescue setting. Together, we demonstrate that de novo VHH design can generate high affinity single-domain antibodies with in vivo protection against lethal cobra venom neurotoxicity, and provide practical insights into method- and framework-dependent performance.

bioengineering↗

bioRxiv2026

Simple Feedback for Complex Movement: Capturing Whole-Limb Reorganization during Single-IMU Gait Retraining

Clinical gait retraining typically relies on multi-sensor arrays and high-dimensional feedback displays, imposing setup and interpretation burdens that limit routine clinical deployment. We developed a single-IMU visual biofeedback system that delivers real-time feedback of Lower Limb Trajectory Error (LLTE), a composite kinematic error metric integrating knee position and shank angle across the stance phase. Twenty able-bodied adults walked on a treadmill under two visual biofeedback targets (flexed-knee, extended-knee) while receiving either corrected (n=10) or uncorrected (n=8) feedback, where the correction accounted for limb orientation at initial contact. LLTE and stance-phase knee kinematics adapted consistently under the flexed-knee target for both feedback groups, with feedback formulation moderating the temporal trajectory of change. Adaptation toward the extended-knee target was limited, likely because participants were already operating near terminal knee extension and because the scalar error metric provided limited directional information for correction. Ankle range of motion (ROM) changed significantly across the stance phase under both target conditions, while hip ROM did not. Multiscale multivariate sample entropy (MSMVSE) increased monotonically with time scale across all conditions, with no statistically distinguishable difference between corrected and uncorrected feedback. These results suggest that single-IMU LLTE biofeedback can modify gait mechanics and that adaptation was expressed across multiple lower-limb segments rather than through changes at a single joint.

bioengineering↗