Nature Machine Intelligence

Accurate, reproducible interpretation of kidney allograft biopsies is critical for diagnosis of graft injury to guide prognosis and management. The international Banff classification is a consensus diagnostic system based on semiquantitative histological lesion scoring on either extent or severity of kidney transplant biopsies. However, pathologist scoring is limited by substantial interobserver variability, constrained scalability, and the inherent nature of the scoring system itself. Here we present BanffNET, a weakly supervised, probabilistic deep learning framework that combines self-supervised feature extraction with a novel Bayesian multiple-instance learning framework to predict (continuously) the full spectrum of Banff lesion scores directly from whole-slide images (WSIs). Using lesion-specific aggregation functions tailored to localized (modeling lesion severity) and diffuse pathologies (modeling lesion extent), BanffNET generates interpretable, patch-level probability maps and calibrated slide-level scores. BanffNET's performance was assessed relative to consensus, biological correlates of rejection and clinical outcome, demonstrating superior consistency, transportability and generalization. Trained on 7,249 WSIs from three cohorts, BanffNET demonstrates consistent performance on 11,028 WSIs across five external test sets, performing on par or exceeding expert consensus across lesions. BanffNET scores align more closely than pathologist Banff scores with molecular profiles of rejection, offering a transparent, biologically grounded framework for computational pathology with relevance beyond transplantation.

Deep learning-based protein structure prediction methods that leverage evolutionary information from multiple sequence alignments (MSAs), exemplified by AlphaFold2, have achieved remarkable accuracy. However, existing methods still struggle to predict challenging proteins, particularly those with novel folds or limited evolutionary information, and to recover alternative conformational states. Here we show that structure prediction models trained under different MSA-depth distributions corresponding to different levels of evolutionary information exhibit complementary generalization behaviors, and that a model trained on a mixture of these distributions can combine their complementary generalization strengths. Building on this insight, we developed ProtMonomer, a deep learning framework trained on MSA-depth distributions representing a broad range of evolutionary information levels to improve structure prediction. Across benchmarks comprising CASP15 targets, non-redundant experimentally determined structures, orphan proteins, and short peptides, ProtMonomer performed comparably to or better than leading methods, including AlphaFold2 and AlphaFold3, with particularly strong performance on challenging targets. For fold-switching proteins, ProtMonomer also recovered alternative conformational states more accurately than AlphaFold2 and AlphaFold3 across diverse homologous sequence sampling strategies. In addition to improving predictive accuracy, ProtMonomer substantially reduced inference cost through an efficient architecture, enabling high-throughput applications. Together, these findings provide insights into the generalization of evolution-informed structure prediction models and support ProtMonomer as an accurate and efficient framework for protein structure prediction.

Deep learning has accelerated drug discovery, yet most existing models are trained using in vitro affinity datasets and consequently remain disconnected from the cellular context in which functional ligand-protein interactions occur. This limitation hinders the ability to reflect the complexity of native interactomes and characterize biological responses to molecular perturbation. Here we introduce C-PLANK (Chemi-Proteome Language Attention NetworK), a deep learning framework trained on fragment-protein interactions profiled directly in living cells using fully functionalized fragment (FFF) chemoproteomics. C-PLANK combines physicochemical embeddings with a bilinear attention network (BAN) to model both global cellular context and local residue-atom interactions, generating interpretable interaction fingerprints. Particularly, C-PLANK incorporates Cellular Interaction State Index (CISI), a systems-level evidential metric that contextualizes the biological plausibility of each predicted interaction against the global cellular interaction landscape. Across 431 ligand interactomes curated from eight independent chemoproteomic studies, C-PLANK consistently outperformed current state-of-the-art interaction prediction frameworks under both random and cold-protein evaluation settings. The inferred interaction fingerprints aligned with orthogonal evidence from structure-based pocket predictions, co-crystal structures, and cellular binding-site annotations. C-PLANK further generalized to unseen ligands. In a cellular target-focused discovery campaign, C-PLANK identified a previously unrecognized ligand that was subsequently advanced into an active chemical probe acting as a SIRT3 agonist in cellular assays. By learning directly from cellular chemoproteomics, C-PLANK moves beyond isolated interaction prediction toward cellular interaction-state modelling, establishing a computational foundation for future digital-twin frameworks in drug discovery.

Protein language models organize sequence and structure at scale, but a global representation of how proteins respond to mutation remains lacking. We present RegimeFormer, a large protein perturbation model coupled to RegimeAtlas, constructed by harmonizing and indexing 202,556,313 non-redundant protein sequences across the tree of life. A diversity-preserving one-million-protein subset provides the high-resolution training and inference layer, with 995,995 proteins yielding residue-level summaries across 407,048,356 residues and substitution-specific predictions available on demand. Across experimental deep mutational scanning, molecular benchmarks, structural confidence and evolutionary constraint, RegimeFormer identifies reproducible protein-level perturbation regimes that organize residue fragility, adaptability and predictive uncertainty. Regime conditioning improves substitution-specific prediction, with the largest relative gains under unseen-protein, unseen-family and low-homology evaluation. RegimeFormer-derived molecular priors further improve downstream transcriptomic and drug-response modelling. Together, RegimeFormer and RegimeAtlas provide a scalable framework for mapping, predicting and querying protein perturbation landscapes across global sequence space.

Recent advances in protein structure prediction, exemplified by AlphaFold, have largely addressed the determination of static structures, one aspect of the protein folding problem. However, predicting folding pathways, by which proteins reach their native states, remains a significant challenge. Here, we present PathFold, a deep learning framework that predicts protein folding pathways directly from sequence information. PathFold leverages an AlphaFold-based module to extract structural information from the sequence and generates a progressive folding trajectory from an extended conformation using a diffusion model. By modeling the full trajectory, it enables prediction of folding intermediates and transition pathways, analogous to those observed in steered molecular dynamics (SMD) simulations. The predicted pathways reveal well-defined intermediates and sequential folding events, and show agreement with experimental folding data, including measured {Phi}-values.

Linker chemistry and conformation are central determinants of PROTAC activity, shaping ternary-complex geometry, cooperativity, target-lysine presentation and cellular permeability. Existing linker generators often lack explicit control over linker flexibility, require predefined attachment sites and linker lengths, or produce structures that demand substantial geometric correction, limiting their utility in practical PROTAC design. Here we introduce FlexiTAC, a Bayesian flow network that jointly generates linker atom types and coordinates from the warhead and E3-ligase-ligand contexts. We also assemble PROTAC-3D, a quality-controlled collection of 63,554 component-resolved PROTAC structures for model training, and PROTAC-Bench, which covers molecular quality, fragment preservation, geometric fidelity, conformational stability, fragment awareness, rediscovery and sampling efficiency. Compared to the best 3D baseline models, FlexiTAC improves validity by 12.0-12.7% and achieves the highest PoseBusters pass rate of 79.5%-80.0%. A differentiable guidance module shifted generated linkers along a conformational ensemble-derived rigidity axis without retraining the generator. In silico case studies further show that the model can accept crystal-derived, redocked or predicted structural inputs. Together, FlexiTAC, PROTAC-3D and PROTAC-Bench establish an integrated and reproducible framework for data-driven PROTAC linker design, combining controllable structure-conditioned generation with standardized training data and evaluation protocols. This framework expands the linker chemical and conformational space accessible to computational exploration, provides a foundation for future method development and enables the systematic generation of structure-conditioned linker designs with tunable conformational flexibility.

Stroke remains a leading cause of mortality and long-term disability worldwide, yet rapid diagnosis is often limited by the shortage of trained radiologists, particularly in resource-constrained settings. Automated analysis of CT imaging offers a potential solution, but existing methods often struggle to achieve clinically generalisable performance while jointly addressing multiple diagnostic tasks. Here we present the Intelligent Integrated Stroke Diagnosis System IISDS, an end-to-end deep learning framework built upon StrokeGNN, a graph-based architecture that integrates 3D contextual feature extraction with U-Net-based 2D lesion segmentation to enable comprehensive stroke analysis from non-contrast CT scans. IISDS performs stroke subtype classification, lesion segmentation and lesion volume estimation within a unified pipeline. To develop and validate the system, we collected and curated BGD-ISD through a collaboration between AI researchers, neurologists, radiologists and clinicians, resulting in a large multi-centre dataset comprising 1,507 CT scans from 597 stroke cases acquired across six hospitals and medical centres in Bangladesh. Across BGD-ISD and multiple publicly available datasets, IISDS achieves state-of-the-art performance on all tasks, improving segmentation accuracy by [≥]0.011 Dice score, reducing lesion volume estimation error by [≥]0.3 average symmetric surface distance (ASSD), and increasing classification performance by [≥]0.018 area under the receiver operating characteristic curve (AUC) compared with existing approaches. These results demonstrate the potential of graph-based deep learning to enable clinically generalisable, automated and scalable stroke diagnosis from CT imaging, supporting rapid clinical decision-making, particularly in healthcare environments with limited access to expert radiological interpretation.

Proteins are dynamic molecules existing in diverse conformational states underlying their biological functions. Although recent approaches have enabled diverse conformational sampling by emulating molecular dynamics simulations, perturbing evolutionary information, or steering internal mechanisms of structure prediction models, predicting conformations resulting from major domain motions or motions that occur over long timescales still remains a challenge. To this end, we introduce GNMCADS, a conformational sampling strategy that enhances the diversity of protein diffusion models by selectively annealing the conditioning signal guided by the intrinsic dynamical organization of the sampled protein. Further, we implement GNMCADS in the diffusion module of AlphaFold3, enabling the generation of diverse protein conformations. When benchmarked across 92 proteins that include 54 class A GPCRs, 15 transporters, and 23 proteins with major domain movements, GNMCADS exhibits improved sampling diversity compared to other current conformational sampling methods.

Improvements in data availability, sharing, and integration, together with the development of explainable artificial intelligence (XAI) techniques, are advancing precision medicine for pediatric cancer by facilitating diagnosis, biomarker discovery, and drug development. Data sharing commons and initiatives like the Childhood Cancer Data Initiative (CCDI) provide access to pediatric-specific genomic and clinical data cohorts and improve data availability for pediatric cancer research. Based on CCDI, a scalable AI platform, Graph Artificial Intelligence for Pediatric Oncology (GAIPO), integrates various data modalities from bulk and single-cell omics data to clinical information. Such multi-modal data facilitates the training and development of advanced XAI models for pediatric cancers. We then developed an end-to-end multi-modality framework, PCGS, for pediatric cancer by incorporating omics-specific representation learning via GNN models with cross-attention fusion and multi-objective learning for downstream tasks such as classification, clustering, and survival analysis. This framework outperforms previous supervised multi-omics integration baseline approaches based on glioma and Wilms tumor cohorts and enables GNN model explainability via Shapley value-based feature attribution approaches to explain the contributions of gene-level features across various biomedical tasks, including classification and survival. Given specific background samples (e.g., age groups, sex, grades) as baselines, this explainable GNN model estimates and ranks the importance scores for input features from each omics modality. It identifies background-specific key features for biomarker discovery, risk group identification, and survival analysis in glioma and Wilms tumor, with potential applicability to other pediatric cancers.

Scaling the structural annotation of protein-coding genes to all eukaryotic genomes remains a major challenge. While recent deep learning methods rival evidence-based pipelines without requiring RNA-seq or alignments, they are entirely supervised. They depend on large, high-quality training sets from diverse genomes, leaving many basal eukaryotic clades without an accurate ab initio gene finder. We present Vipsania, the first unsupervised deep gene finder. A differentiable hidden Markov layer inside a deep sequence model learns to predict gene structures from unannotated genomes alone. Vipsania is pretrained for virtually all eukaryotes and finetunes without supervision on the target genome. It is, on average, more accurate than supervised methods across most clades and avoids the accuracy drop that supervised models suffer on distant target genomes. Vipsania adapts to non-standard genetic codes and provides a fast and highly versatile tool for unbiased, pan-eukaryotic genome annotation. The source code is available at https://github.com/gaius-augustus/vipsania.

Ultrathin endoscopy is highly attractive for real-time tissue imaging in narrow and hard-to-reach regions of the body. A single multimode fibre (MMF) is an attractive probe because of its small diameter, flexibility, and diffraction-limited spatial resolution enabled by the large number of transverse modes guided within a single core. Because the distal fibre tip is inaccessible during endoscopy, reflection-mode imaging, in which the same fibre delivers illumination and collects backscattered light, is more practical than transmission-mode imaging. However, image recovery from the resulting speckle pattern is challenging because light undergoes double-pass propagation through the MMF, with mode coupling and dispersion; the backscattered signal is weak, and the camera records intensity only, without phase information. Here, we propose a single-shot reflection-mode MMF imaging framework that combines a reflected real-valued intensity transmission matrix (reflected-RVITM) with an image restoration network. The reflected-RVITM is calibrated using intensity-only measurements, without interferometry or phase retrieval, and provides a physics-guided initial reconstruction from a single backscattered speckle frame. A restoration network then refines this initial reconstruction instead of inverting the raw speckle. Four restoration backbones are evaluated: HPM-Attention-UNet, GAM, MambaIRv2, and CICPNet. On matched datasets, hybrid models outperformed corresponding networks trained to map raw speckle directly to images. For example, HPM-Attention-UNet on MNIST improved mean PCC from 0.572 to 0.944 (+65.1%). Under domain shift, with training only on Fashion-MNIST and tested on unseen CIFAR scenes, hybrid models achieved mean PCC of 0.61-0.65, compared with 0.36-0.50 for direct learning. This framework is further demonstrated using physical objects at the distal fibre tip. These results demonstrate that a reflected-RVITM physics prior combined with a restoration network enables single-shot image recovery after intensity-only calibration, offering a phase-retrieval-free and generalisable route towards minimally invasive reflection-mode MMF endoscopy.

Spatial omics technologies resolve molecular expression and spatial architecture at single-cell resolution, but profiling whole slides remains costly. In practice, only a few regions of interest (ROIs) are profiled, leaving the rest of the tissue unmeasured. S2-omics was the first framework to unify ROI selection with out-of-ROI prediction, but it operates on superpixels rather than individual cells and predicts discrete cell types rather than continuous molecular profiles. Superpixel-based representations do not explicitly preserve cell boundaries, while categorical cell-type labels cannot quantify molecular expression within cells. Here we present RECON, a two-stage framework that performs ROI inference and whole-slide molecular reconstruction at single-cell resolution, predicting both continuous molecular profiles and discrete cell-type labels. In the first stage, RECON extracts morphological and microenvironmental features from individual cells to identify a representative ROI for spatially resolved single-cell molecular profiling. In the second stage, RECON trains deep learning models on molecular measurements acquired within the selected ROI and reconstructs transcriptomic or proteomic profiles for all remaining cells on the slide. Benchmarked against pathologist annotations, RECONs ROI selection outperforms the superpixel-based S2-omics approaches (IoU: 0.75 versus 0.64). For transcriptomics, refining the modeling unit from superpixels to single cells improves per-gene Pearson correlation by 22%. For proteomics, RECON surpasses the current state-of-the-art method, ROSIE, across all 16 markers, with a median per-cell Pearson correlation of 0.91 versus 0.84. Moreover, RECON delineates tumour boundaries and regions with distinct immune-cell densities, and highlights candidate tertiary lymphoid structures. Together, these results demonstrate that RECON enables informative ROI selection and whole-slide molecular reconstruction at single-cell resolution for both spatial transcriptomics and spatial proteomics.

Label-free Raman imaging of living cells is photon limited: at exposures compatible with cellular dynamics, single-pixel spectra carry about one count per channel on a dominant smooth background. We present an unmixing framework in which the decoder of a physics-constrained autoencoder is restricted to a data-driven spectroscopic dictionary: band centers,widths, and pseudo-Voigt shapes are measured from the dataset and fixed, and the network learns only nonnegative band amplitudes, a smooth B-spline background, and a per-pixel gain.First, on slit-scanning images of HeLa cells (532 nm) the dictionary yields spike-free component spectra that read as band tables, including a resonance-enhanced cytochrome-c-associated component matching literature spectra, and the most stable decomposition against the component number. Second, the dictionary and initialization calibrated at 1 s exposure perline transfer to 100 ms per line (12 s sweeps): cytochrome-c spectral identity survives a single sweep (correlation 0.92) while its map remains photon limited; the dictionary provides spectral physicality, and the transferred initialization prevents a structural collapse that global map correlations miss; in a measurement-derived phantom the dictionary estimator holds thecytochrome-c spectrum to 17-19{degrees} spectral angle at 100 ms, where classical factorizations and free decoders lose it (55-64{degrees}). Estimation on the count-equivalent detector output uses a calibrated shifted-Poisson quasi-likelihood. Third, evaluation must be time matched:correlation against a separately acquired reference saturates through slow specimen drift and acquisition mismatch rather than photon noise, and the self-consistency of learned denoisers is inflated by shared bias; time-matched self-consistency and independent cross-checks areproposed.

Automated lesion segmentation is increasingly central to PSMA PET/CT interpretation, supporting staging, treatment planning, and response assessment at a scale that outpaces available nuclear-medicine expertise. However, automated PSMA-PET/CT whole-body lesion segmentation models are trained on images alone, with no knowledge of where in the body prostate metastases actually tend to occur. Radiologists use clinical domain knowledge of metastatic spread, but its absence in machine learning models produces false positives in anatomically implausible locations and missed lesions in high-risk sites such as the liver. In this work, we explore whether population-level spatial knowledge of metastatic spread can be used to augment deep learning segmentation predictions, and how such a prior should be fused with a network's output, without additional training. We build a data-driven metastasis atlas from 375 expert-annotated whole-body PSMA PET/CT scans and investigate its fusion with a trained segmentation network under a Bayesian framework, in which prediction probabilities from an nnU-Net-based lesion segmentation model serve as the likelihood and the data-driven atlas as the prior. Because metastases occupy only a small fraction of whole-body voxels, the atlas's peak probability is too low, and standard power-scaled or naive Bayesian pooling references lack the tools to deal with this shortcoming. This causes these standard fusion strategies to fail and, in the naive Bayesian case, to sharply degrade performance. We instead derive a calibrated, background-referenced log-odds fusion, one of many possible approaches to combine a population atlas with a deep learning model's predictions, distinct from classical multi-atlas label fusion in that it fuses a single population prior with a trained network's softmax rather than combining several registered atlases. Furthermore, this approach is neutral outside atlas support by construction, reduces exactly to the baseline network when unweighted, and requires no retraining. This atlas fusion significantly improved mean Dice over the baseline nnU-Net on a disjoint internal test set ($+0.011$, Holm-adjusted $p=0.021$) and on an independent external cohort ($+0.0129$, Holm-adjusted $p=3.8\times10^{-16}$), with lesion sensitivity improving from 0.849 to 0.861 internally and Dice improving over baseline in every stratified anatomic region, including the rare, high-risk sites motivating this work, while naive Bayesian pooling degrades performance sharply and power-scaled pooling underperforms it throughout. Our findings suggest that population-level spatial priors can meaningfully augment deep learning predictions in whole-body oncologic segmentation, provided the fusion rule is calibrated to where the prior actually carries signal.

Objective: Published P300-speller fusion schemes fix prior trust regardless of trial reliability; we tested whether a reliability estimate improves on it. Methods: We reanalyzed 3,373 archived P300-speller selections from 47 people with ALS (BigP3BCI). A fair, matched-search-space comparison, tuning both a fixed weight and an adaptive policy out-of-fold, was evaluated across 22 evaluable language-model priors up to 46.7B parameters. Two representative priors, GPT-2 and a classical 5-gram, additionally received detailed naive and mechanistic analyses. Results: No prior's 95% CI favored adaptive fusion under the fair comparison, despite unexploited oracle headroom at every scale. Under GPT-2, the naive comparison was significantly worse for adaptive fusion; both anchors converged to a degenerate or near-degenerate fair-comparison solution. For the representative anchors, three further controllers failed to convert that headroom into benefit; the fixed-fused posterior's output probability outperformed the best controller for flagging errors (2.8- to 3.8-fold enrichment). Conclusion: A tuned fixed weight is a difficult-to-beat default across the tested scale range; reliability estimation gave no deployable adaptive advantage. Significance: Adaptive weighting should be validated against a fairly tuned baseline across model families and scales; in this dataset, the fused output's confidence identified high-risk selections better than the tested purpose-built ranker.

Many biological processes rely on mechanical forces, with protein molecules acting as key mediators. Understanding how proteins respond to mechanical stress is essential for conditions including cardiomyopathy and muscular dystrophy. Natural proteins such as dystrophin and utrophin are composed of heterogeneous folding domains with distinct mechanical properties; deciphering domain-level behavior provides insights into disease mechanisms and informs therapeutic strategies. Single-molecule force spectroscopy (SMFS) enables probing the mechanical properties of entire proteins, yet current approaches struggle to identify heterogeneous folding domains, particularly without prior knowledge. Here, we present the first automated framework to identify heterogeneous folding domains in SMFS data, applying both existing clustering methods and a novel physics-aware deep clustering architecture, LatentUnfold. LatentUnfold learns complementary latent representations from force magnitude and the force-extension physical relationship through dual autoencoders, jointly optimized for clustering assignments. We apply our framework to experimental SMFS data collected from a synthetic two-domain protein (ddFLN4-Titin I27) as well as natural protein constructs of dystrophin and utrophin, with Monte Carlo simulated datasets serving as controlled validation. For the synthetic protein, we recover mechanical properties consistent with previously reported values for each domain. For the natural proteins, we uncover two mechanically distinct domain populations - corresponding to the N-terminal domain and spectrin-like repeats - with differences in both unfolding force and contour length increase, and reveal different unfolding order between them for the first time. This work enables domain-level biological inference, overcoming prior limitations that relied on averaging and overlooked heterogeneity, thus advancing the understanding of mechanical behavior in protein unfolding.

Diffusion models achieve high reconstruction quality in low-dose computed tomography (LDCT), but their iterative sampling trajectories impose substantial computational costs. Unlike unconditional generation, paired LDCT reconstruction starts from an image that already contains the anatomy and spatial structure of the standard-dose CT (SDCT) target; reconstruction primarily requires correcting dose-related noise and artifacts. We therefore introduce Residual Endpoint Flow Matching (REFM), an LDCT reconstruction method that learns to transport an LDCT image directly toward its paired SDCT endpoint rather than defining a noise-to-image trajectory. REFM predicts the residual velocity along linear interpolations between both images and supports single-step and multi-step reconstruction using the same trained network. We evaluate five model capacities using 1 to 50 Euler steps against deterministic U-Net and diffusion-based baselines. Across all REFM variants, one-step inference consistently provides the highest reconstruction quality. On the TCIA validation set, REFM Base achieves 50.98 dB PSNR and 0.9865 SSIM at 94.54 fps, compared with 50.92 dB, 0.9847, and 9.26 fps for DDPM-10. REFM Small retains 50.71 dB while increasing throughput to 198.56 fps. Without fine-tuning, REFM Base also matches the 25-step DDPM baseline on the external Mayo Clinic dataset, although DDPM remains stronger on synthetically degraded CRLM images. Thus, our results show that exploiting paired anatomical correspondence enables diffusion-level LDCT reconstruction with a single step reconstruction.

High-quality experimental datasets that link protein-ligand structures with binding affinity data are essential for developing and evaluating structure-based machine learning methods. To help address this need, we established OpenBind as an open-science initiative to generate large-scale experimental datasets for structure-based AI and molecular discovery. Here, we describe the first public OpenBind release, which, to the best of our knowledge, is the largest public single-target experimental structure-affinity dataset. The dataset focuses on enteroviral 2A protease, comprising 925 crystallographic binding events from 699 compounds and associated affinity measurements for 601 compounds. It combines structures from an initial fragment screen and follow-on molecules, together with affinity data, linking experimentally determined protein-ligand binding modes to biophysical measurements within a coherent antiviral discovery campaign. We used this dataset to evaluate protein-ligand structure prediction, binding-affinity prediction, and virtual screening using representative structure-based methods, including docking and cofolding. This exposed several challenges that are central to practical structure-based modelling: docking performance depends strongly on binding-pocket conformation, poses are difficult to rank, and structure-based affinity prediction remains challenging. Fine-tuning OpenFold3-p2 on the fragment-screen structures substantially improved pose prediction and virtual screening for related follow-on compounds, demonstrating how early-stage experimental structures can support target-specific model adaptation.

SynNotch-CAR circuits improve T cell specificity by coupling antigen recognition to inducible CAR expression. However, basal CAR expression without receptor activation, termed here as leakiness, can reduce the separation between killing of intended target cells and sparing of antigen-positive off-target cells, limiting target-cell discrimination. Here, we systematically quantified basal CAR expression for several synNotch-CAR designs and developed a coupled ordinary differential equation model to show that discrimination depends on basal output, CAR potency, and effector-to-target ratio. We introduced C-terminal tags such as fluorescent proteins, degron domains, endocytosis signals, and endoplasmic reticulum retention motifs as a strategy to reduce CAR leakiness. We found that fluorescent proteins and degron-containing tags reduced basal CAR surface expression while preserving antigen-induced CAR expression, improving discrimination of antigen-density sensing and combinatorial circuits in vitro. In xenograft models, fluorescent protein-tagged CARs improved discrimination by reducing activity against off-target cells while retaining activity against high-antigen tumors. Degron-containing constructs reduced basal CAR expression in vitro but showed suboptimal performance in vivo, revealing a trade-off between basal CAR suppression and induced CAR persistence. Together, these findings demonstrate that basal output expression is a key parameter for inducible genetic circuit designs and establish layered transcriptional and post-translational regulation as a strategy to improve the fidelity of inducible T cell circuits.

Molecular staging of Alzheimer's disease (AD) increasingly defines transition boundaries along single-cell pseudo-progression trajectories, yet whether such boundaries reproduce across brain regions, cohorts and molecular modalities is rarely tested. We present a permutation-controlled audit that combines nine boundary-detection algorithms with a fixed marker panel and four orthogonal reproducibility axes-algorithmic consensus, region, cohort and modality. On synthetic data with planted ground-truth boundaries the audit reaches 100% sensitivity and 94% specificity, rejecting four distinct artefact classes each by a different axis. Applied to the Seattle Alzheimer's Disease Brain Cell Atlas middle temporal gyrus, it localizes a transition that is robust across algorithms and recovered in most cell types but does not generalize: its leading marker is attenuated or absent in prefrontal cortex, entorhinal cortex and cerebrospinal fluid, and an apparent cross-region conservation of glial metabolic genes proves to be a global-expression offset rather than a shared program. The same audit nonetheless certifies an externally validated marker (astrocytic PTGDS) as reproducible across regions and modalities, showing that it separates generalizable anchors from dataset-specific ones rather than rejecting all signals. We provide this four-axis audit as a transferable, code-available standard to apply before a trajectory boundary is read as a biological stage, in AD and other progressive proteinopathies.