Advanced Intelligent Systems

In materials science, the integrity of scanning electron microscopy (SEM) images is paramount for quality control and validation of research outcomes. However, the introduction of sophisticated generative artificial intelligence, particularly Generative Adversarial Networks (GANs), has introduced a novel vulnerability: the potential for highly realistic, artificially synthesized SEM images to be used fraudulently in scientific literature. To address this challenge, we present a deep learning-based framework capable of distinguishing between authentic SEM images and those synthesized by Generative Adversarial Networks (GANs). Using FastGAN and StyleGAN2-ADA, two state-of-the-art GAN models, we generated synthetic SEM datasets to complement real imaging data. We fine-tuned a pre-trained Contrastive Language-Image Pre-training (CLIP) Vision Transformer (ViT-L-14) for binary classification. By unfreezing the final transformer blocks and appending a custom classification head, the model effectively captures the subtle, high-level artifacts inherent in GAN-generated upsampling. This work highlights the potential of deep learning to safeguard scientific imaging workflows and provides an important step toward detecting and mitigating image forgeries in materials science publications.

Controlling cell physiology is difficult, not only because of cells complexity, but also their capacity for real-time adaptation to interventions, leading to challenges such as drug resistance and transgene silencing. Accumulating evidence suggests that this adaptivity resembles classical forms of learning defined in behavioral science. However, a lack of appropriate platforms has led to gaps in our understanding of cells capacity for adaptive problem-solving in physiological and transcriptional space. Here, we present a device, the Cell Trainer, capable of performing a wide variety of automated training experiments on non-neural mammalian cells, using timed drug pulses as the stimulus, and a mobile fluorescence microscope to capture images of responses, across replicate cultures. The Cell Trainer can operate in either an open-loop (feedforward) or closed-loop (feedback-controlled) mode, and our image analysis pipeline can report the behaviors of individual cells throughout each experiment and quantify population heterogeneity. We showcase the ability of the Cell Trainer to execute experimental protocols and perform single-cell analyses in both modes. We first demonstrate with a feedforward experiment in which myoblasts are repeatedly pulsed with dimethyl sulfoxide (DMSO) and their discrete calcium responses are analyzed, revealing sensitization-like dynamics. Next, we demonstrate a feedback control scheme wherein the fluorescence of a pH/voltage reporter in kidney cells is maintained below a threshold level with controlled pulses of acid. To accelerate research in the field of cell training, learning, and memory, we are openly sharing the Cell Trainer schematics and software with the research community. This platform provides a flexible tool for studying how cellular physiological states can be shaped by patterned stimulation and feedback control through approaches that work with the native adaptive competencies of cells.

The lifetime of fluorescent molecules provides an orthogonal readout to fluorescence intensity, opening experimental possibilities of measuring changes in local molecular environments, mechanical tension, and metabolism, among other factors. These changes are best studied live and in vivo; however, limitations of slow imaging speeds, high phototoxicity, and increased data size and complexity have significantly impeded progress on this front. Here, we present a complete and transferable pipeline consisting of a light sheet FLIM microscope and an accompanying machine learning model for data processing that renders long-term and/or high-speed volumetric FLIM (vFLIM) tractable in living systems. We benchmark this pipeline across several biological use cases, model systems, lifetime ranges, and spatiotemporal scales, showcasing a suite of possibilities that our workflow enables. This comprehensive pipeline from imaging to analysis is a crucial step forward towards disseminating the power of live vFLIM to the broader bioimaging community.

The sea anemone Aiptasia is a model system for understanding cnidarian loss of symbiotic algae under heat stress (bleaching). While Aiptasia polyps have been widely used to study this process, accurate symbiosis phenotyping grapples with discordant length scales: fine spatial resolution (~100 um) is needed across a whole organism (~5 mm). To address this, we consider small (~100 um), optically transparent Aiptasia larvae as a bleaching model suitable for whole-organism phenotyping by fluorescence microscopy with larvae classified as symbiotic when algae are localized within gastrodermal cells. To expedite phenotyping, we introduce a machine-learning (ML) image-analysis pipeline (SYMPHONY) designed for single-larva resolution analysis of intact larvae. SYMPHONY efficiently identifies the cellular location of internalized algae (accuracy: 79%, precision: 82%, recall: 79%, F1 score: 79%; training dataset composed of 1611 total objects). Additionally, SYMPHONY reports statistically significant larval bleaching under heat stress and corroborates manual phenotyping results, while significantly reducing operator labor from hours to minutes. The combination of the Aiptasia larvae model and the SYMPHONY pipeline aims to accelerate our understanding of symbiosis breakdown.

Visual neuroscience requires precise spatiotemporal projection of optical stimulation onto the retina, especially in experimental mouse models. However, in vivo patterned stimulation in mice is profoundly hindered by the extreme optical power and severe anatomical aberrations of the eye. Consequently, visual stimulation relies mainly on unverifiable, open-loop approximations that often lack spatial precision. Here, we introduce a closed-loop, spatially modulated stimulation platform that overcomes these barriers. By integrating a digital micromirror device (DMD) with electronically tunable lenses (ETLs) and a real-time, fundus camera-guided focus optimization module, we directly verify the location of patterned stimuli on the retina while dynamically correcting for chromatic and geometric defocus. This platform delivers quantitatively verified static and dynamic patterned stimuli to the living retina with lateral resolutions as fine as 6.7 {micro}m. Guided by ray-tracing optical analysis, our work establishes a technological foundation that enables highly reproducible, cellular-scale interrogations of the visual pathway.

Cell morphology reflects cell health and can distinguish cell-cycle stage, growth arrest, and distinct pathways of cell death. Live, label-free quantitative phase imaging (QPI) captures these features non-invasively and with high temporal resolution, yet many image-based classifiers rely on single frames and cannot separate states whose differences emerge only over time. How much temporal information is needed, and which architecture best exploits it, remain open questions. We assembled 1,874 QPI timelapse sequences spanning six cell states (interphase, mitosis, cell cycle arrest, apoptosis, ferroptosis, and necroptosis) and compared two-dimensional convolutional neural networks (CNNs) with a three-dimensional (3D) spatiotemporal CNN across increasing frame counts. Accuracy improved as frames were added, with the largest gain between one and three frames. The 2D models saturated beyond three frames, whereas the 3D architecture kept improving, reaching 96.5% accuracy and a 3.5% error rate at eleven frames. The temporal information needed tracked the timescale of each process: mitosis was resolved from a single frame, while ferroptosis benefited most from extended sequences. Overall, these results show that dynamic information, rather than static morphology alone, drives accurate cell-state classification, and that 3D architectures are needed to fully exploit it for label-free dynamic phenotyping.

Protocell communities can support programmable molecular nanonetworks, yet most demonstrations use broadcast diffusion or fixed sender-receiver circuits. We introduce PO_SCPLOWROTOC_SCPLOWNO_SCPLOWETC_SCPLOWSO_SCPLOWTACKC_SCPLOW, a network-layer abstraction in which a logical DNA-encoded packet carries a payload, a processing-address list, and an optional forwarding budget. The list determines where localized molecular services transform the packet, not its bidirectional diffusive trajectory. We formulate a finite-state reaction-transport model whose concentration dynamics and single-copy continuous-time Markov chain use the same generator. Under ideal specificity, positive rates, connected transport, no degradation, and sufficient budget, packet stages advance only in the encoded order and delivery occurs almost surely. All injected concentration is delivered asymptotically. Uniform first-order degradation makes delivery probability the Laplace transform of the lossless delivery-time distribution. A union-bound result separates endpoint delivery from route-faithful delivery under off-target processing. As an application, we develop cancellation-based strict-majority aggregation on rooted protocell trees. Conservation of token imbalance proves asymptotic correctness and yields a finite-time certificate. With one initial token per node, outside-root mass below one guarantees the correct root sign. Direct matrix-exponential calculations show sequential processing, branching addressability, route-length attenuation, and bounded forwarding work. A 16-condition finite-copy benchmark with 20,000 trajectories per condition shows that off-target reactions can increase endpoint arrival while decreasing route-faithful delivery. Adaptive ordinary differential equation simulations on trees up to 511 compartments show decision time increasing approximately with maximum tree depth and quantify bias from asymmetric loss. PO_SCPLOWROTOC_SCPLOWNO_SCPLOWETC_SCPLOWSO_SCPLOWTACKC_SCPLOW is therefore a formally analyzable molecular networking architecture and an experimentally testable blueprint. Sequence-resolved gates and chassis calibration remain future work.

While light-sheet microscopy is emerging as a robust method for volumetric imaging with improved axial resolution, its capability regarding two-dimensional, surface-level mapping is often hindered by limitations in data redundancy and reconstruction efficiency stemming from volumetric registration methods. We demonstrate that a multiview imaging approach in an axially-swept, dithered light-sheet microscope paired with computational image reconstruction of view projections is able to address these trade-offs to enable large-scale mapping of surface structural features, leveraging the advantages of multiview light-sheet in scalable field of view, working distance, and near isotropic resolution across the entire imaging depth. To aid in the acquisition and analysis of two-dimensional surface structures, we present a tailored surface mapping workflow and a Fiji plugin for computational reconstruction, promoting robust and comprehensive visualization of surface features of uncleared volumetric samples. Our strategy, termed projection reconstruction for imaging surface morphology (PRISM), integrates axially swept dithered light-sheet microscopy and post-processing software for multiview imaging. The imaging hardware enables near-isotropic resolution across its entire field of view, while the software implementation leverages rigid and affine transformations to align two-dimensional projections of multiview samples. It is designed to work with the BigStitcher pipeline, leveraging its robust algorithm to provide support for two-dimensional image alignment and stitching. We demonstrate the capability of PRISM in studies of lymphatic network mapping in the epicardial layer of intact mouse hearts, as well as surface profiles of FaDu spheroids labeled with antibody-nanodiamond conjugates. This method allows us to quantify cardiac lymphatic branch numbers, diameters, and lengths of a Prox1-tdTomato mouse cardiac model, as well as cluster number and diameters of epidermal growth factor receptor within a FaDu spheroid labeled with a nanodiamond-antibody conjugate, with a significant reduction of post-processing data size. PRISM leverages multiview image projections to promote studies of cardiac lymphatics in mouse models and surface receptor distributions within spheroid models, enabling efficient surface mapping of large, intact, and uncleared biological samples across a variety of scales.

Electrostatic actuation is an emerging technology for generating tactile sensations on capacitive touchscreens through voltage-induced attractive forces between a fingertip and the surface. However, accurate control of electrostatic attraction during natural touchscreen interactions remains challenging because the applied normal force and sliding speed continuously vary, and their effects on the fingertip-screen contact and resulting actuation strength are not fully characterized. Here, we show how normal force and sliding speed systematically alter fingertip- screen contact area and electrical impedance, and use these measured changes to estimate electrostatic attraction during sliding. Contact area, interaction forces, and electrical impedance were measured simultaneously as participants slid their fingertips across an electrostatic surface under systematically varied normal forces and sliding speeds. These measurements revealed condition-dependent changes in fingertip contact, electrical interaction impedance, effective capacitance, derived effective gap thickness, and electrostatic attraction. We then incorporated these measured contact quantities into a physics-informed, data-driven model based on parallel-plate capacitor theory, in which effective capacitance, apparent contact area, and effective voltage determine the estimated electrostatic attraction. The resulting model links force- and speed-dependent changes in these quantities to electrostatic attraction while accounting for inter-participant variability through a participant-specific scaling factor. These findings provide experimentally grounded guidance for designing electrostatic surface-haptic feedback and future adaptive control strategies under realistic touch conditions.

Cardiac blood flow is a regulator of several important developmental and remodelling processes in the heart, including through fluid shear forces sensed by the endothelial cells lining the heart. However, optically mapping these flow fields in the complex 3D geometry of the heart is challenging even in transparent animal models such as the zebrafish. One of the main challenges is the difficulty in measuring the out-of-plane (axial) velocity component, preventing accurate mapping of the complete 3-component-3-dimension (3C-3D) blood flow velocity field; image-based techniques such as microscopic particle image velocimetry ({micro}PIV) traditionally only provide the in-plane flow components. Here we present a computational approach to achieve full time-varying 3C-3D blood flow vector mapping using a standard selective plane illumination microscope (SPIM), based on robust cardiac phase assignment, precise measurement-driven registration of sequentially acquired z-stacks, and PIV data fusion from multiple sample orientations. Our approach holds the key to understanding the complex dynamic flow fields within the developing heart, and their role in shaping cardiac development.

DNA data storage offers exceptional density and millennial-scale stability, with advances in encoding schemes and reduced synthesis costs making large-scale archiving increasingly viable. However, while efforts have focused on reliable data retrieval, securing DNA-encoded information against unauthorized access remains largely unexplored. Here, we introduce DNA-GUARD (DNA Gated Unlocking and Access Restriction of Data), a molecular-level access control system that physically restricts data retrieval rather than relying on computational encryption. DNA-GUARD integrates with PCR-based random access by selectively blocking amplification of protected sequences. Chemically modified "locker strands" outcompete PCR primers and block polymerase extension through 3 inverted dT modifications, preventing amplification of key sequences required for file decoding. To restore access, complementary "password strands" tethered to magnetic particles sequester locker strands, enabling their removal and restoring data access. We demonstrate DNA-GUARDs scalability from 550-byte to 1-MB files without performance loss, orthogonal control of multiple files within mixed libraries, and reliable repeated locking-unlocking cycles. This approach enables physical access control compatible with established DNA storage workflows, providing a foundation for secure archival storage with implications for molecular information security that complements cryptographic data protection methods.

Understanding how cells establish spatial organization within tissues is a fundamental question in life sciences. While modern three-dimensional fluorescence microscopy captures large-volume tissue architecture, extracting quantitative cellular insights from complex volumetric datasets remains a major barrier. Here, we introduce FOCUS-3D, a robust, broadly generalizable volumetric cell segmentation framework built on a large, diverse manually annotated cell resource and advanced AI designs. Integrating volumetric representation learning, multi-scale feature extraction, and query-based mask prediction, FOCUS-3D achieves state-of-the-art performance across diverse species, tissues, fluorescent reporters and imaging modalities. During zebrafish (Danio rerio) development, FOCUS-3D uncovers three successive phases of notochord morphogenesis. We disentangle early motility-driven rearrangements from later cell shape remodeling and tissue repacking, and further link these morphological states to spatial and developmental transcriptional programs across independent datasets.

High-speed three-dimensional imaging in scattering tissues remains challenging because volumetric microscopy generally requires scanning, whereas snapshot light-field approaches divide limited detector pixels among multiple views. This constraint is particularly severe in the second near-infrared window (NIR-II), where commonly used InGaAs cameras typically have relatively small sensor formats and high detector noise. Here we introduce NIR-II squeezed light-field microscopy (NIR-II SLIM), which optically rotates and compresses multiple perspective views before detection, allowing efficient use of camera pixels while retaining complementary spatial information for three-dimensional reconstruction. NIR-II SLIM acquires volumes at up to 600 volumes s-1 with a reconstructed lateral sampling grid of 512 x 512 pixels. We use the method for label-free four-dimensional imaging of cardiac dynamics in pigmented late-larval zebrafish, resolving chamber deformation and millisecond-scale atrioventricular-valve motion, and for NIR-II fluorescence imaging of vascular and lymphatic transport in mice. NIR-II SLIM provides a detector-efficient approach for high-speed volumetric imaging of rapid biological dynamics in scattering tissues.

Fingertip friction plays a central role in tactile exploration and object manipulation. During sliding, tangential force depends jointly on the real contact area and the interfacial shear stress, both of which can be influenced by sliding conditions. However, changes in fingertip friction are often interpreted primarily through changes in real contact area, whereas the accompanying changes in interfacial shear stress remain less well characterized. This gap is especially relevant for electrostatic surface haptic displays, which modulate fingertip friction by applying a voltage between the finger and the touch surface. Here, we experimentally quantify the mean interfacial shear stress of a sliding fingertip on an electrostatically actuated touchscreen using simultaneous measurements of tangential force and optically resolved real contact area. Ten participants performed sliding trials across three speeds and three normal forces with and without electrostatic actuation. Interfacial shear stress increased with speed and decreased with normal force; in both cases, these trends arose because real contact area varied more strongly than tangential force. Electrostatic actuation further reduced interfacial shear stress, as increasing voltage produced a larger increase in real contact area than in tangential force. These findings show that interfacial shear stress varies systematically with sliding conditions and electrostatic actuation, clarifying how changes in real contact area and interfacial shear stress combine to shape fingertip-surface friction.

Modern biological imaging generates large, complex datasets that require scalable and reproducible image analysis methods. Deep learning has demonstrated strong performance on bioimage segmentation tasks, but training custom models has remained inaccessible to many researchers due to requirements for GPU infrastructure, programming expertise, and large annotated training datasets. ZEISS arivis Cloud is a browser-based platform for deep learning model training that addresses these barriers through partial annotation support, AI-assisted labeling with SAM (Segment Anything Model), pretrained model initialization, and automatically configured training pipelines requiring no machine learning expertise. The platform supports two segmentation tasks: semantic segmentation using a U-Net-style architecture with an EfficientNet encoder and PixelShuffle decoder, and instance segmentation based on Mask2Former with a Swin-Tiny backbone. Both pipelines incorporate microscopy-specific adaptations including smooth tiling, multi-channel input support, dataset-specific normalization, and partial-annotation-aware loss functions protected by patents US-20240078681-A1 and US-20250111519-A1. Trained models integrate directly with ZEISS arivis Pro for pipeline-based image analysis, ZEISS arivis Hub for parallel execution across large datasets, and ZEISS ZEN for content-aware guided acquisition. We describe the platform architecture, training methodology, segmentation architectures, reproducibility and versioning mechanisms, and FAIR compliance, and illustrate the complete workflow through two intestinal organoid imaging examples. arivis Cloud is freely accessible to student users; other users access the platform via subscription at https://www.arivis.cloud/.

While geometric constraints shape tissue development, quantifying the resulting growth dynamics remains a central challenge in tissue engineering. Conventional methods often struggle to capture multi-scale kinetics without complex labeling or difficult single-cell tracking. Here, we analyze geometrically controlled growth of microtissues derived from human dermal fibroblasts using time-resolved, label-free brightfield microscopy, combined with optical flow and semi-automated deep learning mitosis detection. By extracting multi-scale flow fields and integrating them with tissue segmentation, we quantify directional tissue dynamics, separating flow into parallel and normal components relative to the local tissue contour. Applying this framework, we contrast the quiescent tissue interior with the advancing growth front where localized dynamics and cell proliferation drive expansion. Our results demonstrate that, compared to the bulk, the growth front exhibits higher fluctuations parallel to the tissue contour, positive mean normal flow, and significantly increased mitotic activity. Furthermore, evaluating flow divergence around mitotic events reveals distinct spatial behaviors: with the onset of mitosis, a contraction and subsequent expansion occurs in the vicinity of the dividing cells. Beyond the immediate cellular neighborhood, the broader regional dynamics remain consistent before and after mitosis onset, with net tissue expansion in proximity to the growth front and contraction within the tissue interior. By extracting continuous kinetic data from easily accessible, label-free brightfield imaging, this approach serves as a non-invasive, complementary tool for evaluating in vitro tissue morphogenesis and growth dynamics. This analytical framework can be expanded to study locally resolved tissue morphogenesis and growth kinetics in other microsystems, ranging from embryos to organoids. Statement of SignificanceUnderstanding how localized cellular forces drive tissue growth is critical for mechanobiology. However, mapping these dynamics traditionally requires complex, invasive fluorescent labeling. We present an accessible, label-free computational framework combining optical flow and deep learning-based mitosis detection to quantify continuous tissue kinematics directly from standard brightfield microscopy. Applying this to 3D microtissues, we reveal a distinct spatial coupling between cell division, local mechanical fluctuations, and directed tissue expansion at the active growth front. This non-invasive approach bridges the gap between single-cell mechanics and macroscopic morphogenesis, offering a versatile tool to monitor complex in vitro model systems-like organoids and bioengineered tissues-without disrupting their native state.

Host-parasite interactions are dynamic systems, where parasites usually outnumber hosts by one or more orders of magnitude. However, our understanding is often limited by the assessment of parts of the host, at arbitrary time points, and/or aggregate parasite responses. Here we combined custom-built 3D-printed hardware and deep-learning-based algorithms to enable holistic (i.e. all infecting individuals on the whole plant), spatio-temporal, and parasite-centric analyses of plant-parasitism by nematodes from timelapse videos of infection over months. In so doing, we tracked the dynamic growth and development of all individual parasites, at the organismal level, for thousands of hosts across hundreds of genotypes of Arabidopsis thaliana. Categorising traits into the static (i.e. in an acquired image at a given time point) and dynamic (i.e. phenotypic changes over time), we revealed a greater extent of host-genetic control of parasite traits, and new physiological limits of the species under these conditions. Using this capability, we identify Quantitative Trait Loci (QTL) in the host plant associated with 18 phenotypic traits in the parasite as a resource for the community. Finally, we leverage the large and diverse dataset to understand fundamental features of the parasite, independent of host genotype, revealing aspects of the life cycle which are pseudo-deterministic as well as local, deleterious interactions between co-infecting parasites. Given that plant-parasitic nematodes cause an estimated $100 billion in agricultural damages per year, these insights are contextualised in a global challenge driven by plant-parasitic nematodes.

Deep-learning-based segmentation algorithms have gained considerable accuracy for processing biological images. In particular, the introduction of large foundation models, novel architectures, and semantically varied datasets now allows for deployment of state-of-the-art models for clean image cohorts with limited re-training or, in the best of cases, in an out-of-the-box fashion. Biological imaging, however, is liable to corruptions that can hinder their deployment. While some methods document their robustness to the most common corruptions, a systematic robustness analysis of the state of the art to the expansive gamut of corruptions in biological imaging remains to be done. We perform this benchmarking by simulating 36 corruption types with varying degradation severity on images sampled from 30 different datasets. Our benchmark accounts both for the variety in biological images and the nature of corruptions. Among other things, our study reveals that performance on clean images does not correlate with overall robustness to image corruptions. In fact, we find that a decade-old method, StarDist, is more robust than many of its more recent foundation-model-based counterparts. We also show in a dedicated representation analysis that the performance of segmentation models collapses in the early layers of the encoding phase.

Spectroscopic single-molecule localization microscopy (sSMLM) enables multiplexed super-resolution imaging by simultaneously acquiring the spatial position and spectral information of individual fluorophores. Dual-wedge prism (DWP)-based implementations provide a compact, alignment-stable approach to spectral dispersion, but trade-offs between localization precision, spectral precision, and experimental complexity remain. We systematically compare five DWP-based sSMLM configurations, including two-dimensional (2D) and three-dimensional (3D) implementations using single DWP (DWP-sSMLM) and symmetrically-dispersed DWP (SDDWP-sSMLM). We evaluate lateral precision, spectral precision, and ease of use. SDDWP configurations acquire spectral images in both channels and utilize both for spatial localization, yielding the highest lateral and spectral precision. However, for applications that do not require axial information, 2D-DWP provides a simple, plug-and-play solution with robust performance. This work offers a guideline for selecting DWP configurations based on experimental needs.

Quantitative analysis of single-molecule localisation microscopy (SMLM) data remains challenging because biologically diverse, well-annotated datasets are limited, whilst nanoscale protein organisation is heterogeneous and difficult to describe with hand-tuned metrics. We present SynthMLM, a framework that infers interpretable structural descriptors from experimental SMLM data and uses these descriptors to generate synthetic localisation datasets. We demonstrate SynthMLM by generating descriptor-matched synthetic datasets corresponding to diverse experimental SMLM datasets and evaluating their agreement with real data using descriptor-level and embedding-based measures. By enabling controlled generation of synthetic localisation data, SynthMLM provides a practical resource for benchmarking SMLM analysis methods, testing algorithm failure modes, and developing machine-learning workflows where large, labelled datasets are required.