Nature Machine Intelligence

Gene networks (GNs) encode diverse molecular relationships and are central to interpreting cellular function and disease. The heterogeneity of interaction types has led to computational methods specialized for particular network contexts. Large language models (LLMs) offer a unified, language-based formulation of GN inference by leveraging biological knowledge from large-scale text corpora, yet their effectiveness remains sensitive to prompt design. Here, we introduce Gene-Relation Adaptive Soft Prompt (GRASP), a parameter-efficient and trainable framework that conditions inference on each gene pair through only three virtual tokens. Using factorized gene-specific and relation-aware components, GRASP learns to map each pair's biological context into compact soft prompts that combine pair-specific signals with shared interaction patterns. Across diverse GN inference tasks, GRASP consistently outperforms alternative prompting strategies. It also shows a stronger ability to recover unannotated interactions from synthetic negative sets, suggesting its capacity to identify biologically meaningful relationships beyond existing databases. Together, these results establish GRASP as a scalable and generalizable prompting framework for LLM-based GN inference.

Machine learning models trained on observational data from one environment frequently fail when deployed in another, because standard learning algorithms exploit spurious correlations alongside causal ones. Invariant learning methods address this problem by seeking representations that support stable prediction across training environments, but their behavior on tabular data remains poorly characterized. We present CausTab, a gradient variance regularization framework for causal invariant representation learning on mixed tabular data. CausTab penalizes the variance of parameter gradients across training environments, providing a richer invariance signal than the scalar penalty used by Invariant Risk Minimization (IRM). We provide formal results showing that the gradient variance penalty is zero at causally invariant solutions and positive at solutions that rely on spurious features. Through experiments on synthetic data across three spurious-correlation regimes, four cycles of the National Health and Nutrition Examination Survey (NHANES), and four hospital systems in the UCI Heart Disease dataset, we demonstrate that: (1) IRM consistently degrades relative to standard empirical risk minimization (ERM) on tabular data, losing up to 13.8 AUC points in spurious-dominant settings, a failure we trace mechanistically to penalty collapse during training; (2) CausTab matches or exceeds ERM in every experimental condition; (3) CausTab achieves consistently better probability calibration than both ERM and IRM; and (4) invariant learning methods fail when environments differ in outcome prevalence rather than in spurious feature correlations, a boundary condition we characterize both empirically and theoretically. We introduce the Spurious Dominance Index (SDI), a practical scalar diagnostic for determining whether a dataset requires invariant learning, and validate it across all experimental settings

The apolipoprotein E {varepsilon}4 (APOE4) allele is the strongest genetic risk factor for late-onset Alzheimer's disease (AD), the most common form of dementia. APOE4 carriers exhibit cerebrovascular and metabolic dysfunction, structural brain alterations, and gut microbiome changes decades before the onset of clinical symptoms. A better understanding of the early manifestation of these physiological changes is critical for the development of timely AD interventions and risk reduction protocols. Multimodal datasets encompassing a wide range of APOE4- and AD-associated biomarkers provide a valuable opportunity to gain insight into the APOE4 phenotype; however, these datasets often present analytical challenges due to small sample sizes and high heterogeneity. Here, we propose a two-stage multimodal AI model (APOEFormer) that integrates blood metabolites, brain vascular and structural MRI, microbiome profiles, and other clinical and demographic data to predict APOE4 allele status. In the first stage, modality-specific encoders are used to generate initial representations of input data modalities, which are aligned in a shared latent space via self-supervised contrastive learning during pretraining. This objective encourages the learning of informative and consistent representations across modalities by leveraging cross-modality relationships. In the second stage, the pretrained representations are used as inputs to a multimodal transformer that integrates information across modalities to predict a key AD risk genetic variant (APOE4). Across 10 independent experimental runs with different train-validation-test splits, APOEFormer predicts whether an individual carries an APOE4 allele with an average accuracy of 75%, demonstrating robust performance under limited sample sizes. Post hoc perturbation analysis of the predictive model revealed valuable insights into the driving components of the APOE4 phenotype, including key blood biomarkers and brain regions strongly associated with APOE4.

Obtaining novel antibodies against specific protein targets is a widely important yet experimentally laborious process. Meanwhile, computational methods for antibody design have been limited by low success rates that currently require resource-intensive screening. Here, we introduce Germinal, a broadly enabling generative pipeline that designs antibodies against specific epitopes with nanomolar binding affinities while requiring only low-n experimental testing. Our method co-optimizes antibody structure and sequence by integrating a structure predictor with an antibody-specific protein language model to perform de novo design of functional complementarity-determining regions (CDRs) onto a user-specified structural framework. When tested against four diverse protein targets, Germinal successfully designed functional antibodies across all targets and binder formats, testing only 43-101 designs for each antigen. Validated designs also exhibited robust expression in mammalian cells and high sequence and structural novelty. We provide open-source code and full computational and experimental protocols to facilitate wide adoption. Germinal represents a milestone in efficient, epitope-targeted de novo antibody design, with notable implications for the development of molecular tools and therapeutics.

Background. Federated learning (FL) enables collaborative model training across institutions without sharing patient-level data. However, standard FL algorithms such as FedAvg degrade under non-independently and non-identically distributed (non-IID) data, a prevalent condition when patient demographics, scanner hardware, and disease prevalence differ across hospital sites. Objective. We propose iPS-MFFL (Individualized Per-Site Meta-Federated Feature Learning), a federated framework with a hierarchical local-model architecture that addresses non-IID heterogeneity through (1) a shared feature extractor, (2) multiple weak-learner classification heads that can be trained with heterogeneous training objectives to promote complementary decision boundaries, (3) independent per-learner server aggregation so that each weak learner's parameters are averaged only with its counterparts at other clients, and (4) a lightweight meta-model, itself federated, that adaptively stacks the weak-learner outputs. Methods. We evaluate on the Brain Tumor MRI Classification dataset (7,200 images; 4 classes: glioma, meningioma, pituitary tumor, no tumor) partitioned across K = 5 simulated hospital sites using Dirichlet non-IID sampling (alpha = 0.3). Four baselines are compared: Local-only training, FedAvg, FedProx, and Freeze-FT. All experiments are repeated over three random seeds (13, 42, 2025) and evaluated using paired t-tests, Cohen's d effect sizes, and post-hoc power analysis.

Background: Patient-ventilator synchrony, an essential prerequisite for non-invasive mechanical ventilation, requires an accurate matching of every phase of the respiration between patient and the ventilator. Methods: We developed a long short-term memory (LSTM)-based model that can predict the inspiratory and expiratory time of the patient. This model consisted of two hidden layers, each with eight LSTM units, and was trained using a dataset of approximately 27000 of 500-ms-long flow signals that captured both inspiratory and expiratory events. Results: The LSTM model achieved 97% accuracy and F1 score in the test data, and the average trigger error was less than 2.20%. In the first trial, 10 volunteers were enrolled. In "Compliance" mode, 78.6% of the triggering by the LSTM model was compatible with neuronal respiration, which was higher than Auto-Trak model (74.2%). Auto-Trak model performed marginally better in the modes of pressure support = 5 and 10 cmH2O. Considering the success in the first clinical trial, we further tested the models by including five patients with acute respiratory distress syndrome (ARDS). The LSTM model exhibited 60.6% of the triggering in the 33%-box, which is better than 49.0% of Auto-Trak model. And the PVI index of the LSTM model was significantly less than Auto-Trak model (36.5% vs 52.9%). Conclusions: Overall, the LSTM model performed comparable to, or even better than, Auto-Trak model in both latency and PVI index. While other mathematical models have been developed, our model was effectively embedded in the chip to control the triggering of ventilator. Trial registration: Approval Number: 2023ZDSYLL348-P01; Approval Date: 28/09/2023. Clinical Trial Registration Number: ChiCTR2500097446; Registration Date: 19/02/2025.

Background: Paediatric pneumonia is a leading cause of childhood morbidity and mortality worldwide. Chest X-rays (CXR) are an important diagnostic tool in the diagnosis of pneumonia, but shortages in specialist radiology services lead to clinically significant delays in CXR reporting. The ability to communicate findings both to clinicians and laypersons allows MLLMs to be deployed throughout clinical workflows, from image analysis to patient communication. However, MLLMs currently underperform state-of-the-art deep learning classifiers. Objective: To evaluate the diagnostic accuracy of ensemble strategies with MLLMs compared to the baseline average agent for paediatric radiological pneumonia detection. Methods: We conducted a retrospective cohort study using paediatric CXRs from two independent hospital datasets totalling 2300 CXRs. Fifteen MedGemma-4B-it agents independently classified each CXR into five pneumonia likelihood categories. Majority voting, soft voting, and GPTOSS-20B aggregation were compared against the average agent performance. The primary metric evaluated was OvR AUROC. Secondary metrics included accuracy, sensitivity, specificity, F1-score, Cohen's kappa, and OvO AUROC. Results: Soft voting achieved improvements in OvR AUROC (p_balanced = 0.0002, p_real-world = 0.0003), accuracy (p_balanced = 0.0008, p_real-world < 0.0001), Cohen's Kappa (p_balanced = 0.0006, p_real-world = 0.0054) and OvO AUROC (p_balanced < 0.0001, p_real-world = 0.0011) across both datasets, and a superior F1-value (pbalanced = 0.0028) for the balanced dataset. Conclusion: Soft voting enhances MedGemma's diagnostic discriminatory performance for paediatric radiological pneumonia detection. Our system enables privacy-preserving, near real-time clinical decision support with explainable outputs, having potential for integration into emergency departments. Our system's high specificity supports triage by flagging high-risk radiological pneumonia cases.

Automating Hierarchical Condition Category (HCC) assignment directly from unstructured electronic health record (EHR) notes remains an important but understudied problem in clinical informatics. We present HCC-Coder, an end to end NLP system that maps narrative documentation to 115 Centers for Medicare & Medicaid Services(CMS) HCC codes in a multi-label setting. On the test dataset, HCC-Coder achieves a macro-F1 of 0.779 and a micro-F1 of 0.756, with a macro-sensitivity of 0.819 and macro-specificity of 0.998. By contrast, Generative Pre-trained Transformer (GPT)-4o achieves highest score of a macro-F1 of 0.735 and a micro-F1 of 0.708 under five-shot prompting. The fine-tuned model demonstrates consistent absolute improvements of 4%-5% in F1-scores over GPT-4o. To address severe label imbalance, we incorporate inverse-frequency weighting and per-label threshold calibration. These findings suggest that domain-adapted transformers provide more balanced and reliable performance than prompt-based large language models for hierarchical clinical coding and risk adjustment.

Large language models are increasingly used for clinical guidance while their parent companies introduce advertising. We tested whether pharmaceutical ads embedded in the prompts of 12 models from OpenAI, Anthropic, and Google shift drug recommendations across 258,660 API calls and four experiments probing distinct epistemic conditions. When two drugs were both guideline appropriate, advertising shifted selection of the advertised drug by +12.7 percentage points (P < 0.001), with some model scenario pairs shifting from 0% to 100%. Google models were the most susceptible (+29.8 pp), followed by OpenAI (+10.9 pp), while Anthropic models showed minimal change (+2.0 pp). When the advertised product lacked evidence or was clinically suboptimal, models resisted. This reveals a structured vulnerability: advertising does not override medical knowledge but fills the space where clinical evidence is underdetermined. An open response sub analysis (2,340 calls across three representative models) confirmed that advertising restructures free-text clinical reasoning: models echoed ad claims at 2.7 times the baseline rate while maintaining high stated confidence and rarely disclosing the ad. Susceptibility was provider dependent (Google: +29.8 pp; OpenAI: +10.9 pp; Anthropic: +2.0 pp). Because this bias operates within clinically correct answers, it is invisible to accuracy based evaluation, identifying a class of AI safety vulnerability that standard testing cannot detect.

Objective detection of evoked potentials (EPs) is central to digital diagnostics in hearing assessment and clinical neurophysiology, yet current approaches remain time-intensive and sensitive to inter-individual noise variability. Many existing detection methods rely on population-based assumptions or computationally demanding procedures, limiting robustness and efficiency in real-world clinical settings. We present Fmpi, a digital EP detection framework enabling individualised, real-time response detection through analytical modelling of the spectral colour and temporal dynamics of background noise within each recording. Using extensive simulations and large-scale human electroencephalography datasets spanning brainstem, steady-state, and cortical EPs recorded in adults and infants, we demonstrate performance comparable or superior to state-of-the-art bootstrapped methods while operating at a fraction of the computational cost and maintaining well-controlled sensitivity with improved specificity. Importantly, Fmpi incorporates a futility detection mechanism enabling early termination of uninformative recordings, reducing testing time without compromising diagnostic reliability.

Long COVID affects millions of people worldwide, yet no disease-modifying treatment has been approved, and existing interventions have shown only modest and inconsistent benefits. A key reason for this limited progress is that current computational drug repurposing pipelines do not match well with the clinical reality of Long COVID. These patients often have persistent, multisystemic symptoms and may already be taking multiple medications, making treatment safety a primary concern. However, most repurposing workflows still treat safety as a downstream filter and rely on disease-associated targets rather than causal drivers. They also assume that the findings of one analysis would generalize across the diverse presentations of Long COVID. We introduce SPLIT, a safety-first repurposing framework that addresses these limitations. SPLIT prioritizes safety at the start of the candidate evaluation, integrates complementary causal inference strategies to identify likely driver genes, and uses a counterfactual substitution design to compare drugs within specific cohort contexts. When applied to cognitive and respiratory Long COVID cohorts, SPLIT revealed three main findings. First, drugs with similar predicted efficacy could have very different predicted safety profiles. Second, the drugs flagged as unfavorable were often different between the two cohorts, showing that drug prioritization is phenotype-specific. Third, SPLIT flagged 18 drugs currently under active investigation in Long COVID trials as having unfavorable predicted profiles. SPLIT provides a practical framework to identify safer, more context-appropriate candidates earlier in the process, supporting more targeted and better-tolerated treatment strategies for Long COVID.

Predicting response to induction chemotherapy (IC) and overall survival (OS) is critical for optimizing treatment in patients with locally advanced nasopharyngeal carcinoma (LANPC). This study aimed to develop and validate a multi-task deep learning model integrating pretreatment MRI and whole slide images (WSIs) to predict IC response and OS in LANPC. Pretreatment MRI and WSIs from 404 patients with LANPC were retrospectively collected to construct a multi-task model (MoEMIL) for the simultaneous prediction of early IC response and OS. MoEMIL employed multi-instance learning to process WSIs, PyRadiomics and a convolutional neural network (ResNet50) to extract MRI features, and fused multimodal features through a multi-gate mixture-of-experts architecture. Clustering-constrained attention multiple instance learning and gradient-weighted class activation mapping were applied for visualization and interpretation. MoEMIL effectively stratified patients into good and poor IC response groups, achieving areas under the curve of 0.917, 0.869, and 0.801 in the train, validation, and test sets, respectively, and outperformed the deep learning radiomics model, the pathomics model and TNM staging. The model also stratified patients into high- and low-risk OS groups (P < 0.05). MoEMIL shows promise as a decision-support tool for early IC response prediction and prognostication in LANPC. Author SummaryWe have developed a deep learning model that integrates two types of medical images, including magnetic resonance imaging (MRI) and digital pathological slices, to simultaneously predict response to induction chemotherapy and prognosis in patients with locally advanced nasopharyngeal carcinoma. Current treatment decisions primarily rely on traditional tumor staging (TNM), which often fails to comprehensively reflect the complexity of the disease. Our model, named MoEMIL, was trained and tested on data from 404 patients across two hospitals and consistently outperformed both single-model approaches and TNM staging methods. By identifying patients who exhibit poor response to induction chemotherapy or higher prognostic risk, our tool can assist clinicians in achieving personalized treatment, enabling intensified management for high-risk patients and avoiding unnecessary side effects for low-risk patients. Additionally, we visualize the models reasoning process through heat map generation, which highlights the image regions exerting the greatest influence on prediction outcomes. This work represents a step toward more precise treatment for nasopharyngeal carcinoma; however, larger-scale prospective studies are required before the model can be integrated into routine clinical practice.

Background. Pancreatic ductal adenocarcinoma (PDAC) has a five-year survival rate of approximately 12%, largely because it is typically diagnosed at an advanced stage. CT-based computational methods for early detection exist but rely on black-box deep learning or large texture feature sets without tissue-specific interpretability. Methods. We developed Virtual Spectral Decomposition (VSD), which applies six parameterized sigmoid functions S(HU) = 1/(1+exp(-alpha x (HU - mu))) to standard portal-venous CT, decomposing each pixel into tissue-specific response channels for fat (mu=-60), fluid (mu=10), parenchyma (mu=45), stroma (mu=75), vascular (mu=130), and calcification (mu=250). Dendritic Binary Gating identifies structural content per channel using morphological filtering, enabling co-firing analysis and lone firer identification. A 25-feature signature was extracted per patient. Three independent datasets were analyzed: NIH Pancreas-CT (n=78 healthy), Medical Segmentation Decathlon Task07 (n=281 PDAC, paired tumor/adjacent tissue), and CPTAC-PDA from The Cancer Imaging Archive (n=82, multi-institutional, with DICOM time point tags). The same six sigmoid parameters were used across all datasets without retraining. Results. VSD achieved AUC 0.943 for field effect detection (healthy vs cancer-adjacent parenchyma) and AUC 0.931 for patient-stratified tumor specification on MSD. On CPTAC-PDA, VSD achieved AUC 0.961 (6 features) and 0.979 (25 features) for distinguishing healthy from cancer-bearing pancreas on scans obtained prior to pathological diagnosis. All significant features replicated across datasets in the same direction: z_fat (d=-2.10, p=3.5e-27), z_fluid (d=-2.76, p=2.4e-38), fire_fat (d=+2.18, p=1.2e-28). Critically, VSD severity did not correlate with days-from-diagnosis (r=-0.008, p=0.944) across a range of day -1394 to day +249. Patient C3N-01375, scanned 3.8 years before pathological diagnosis, had VSD severity 1.87, well above the healthy mean of 0.94 +/- 0.33. The tissue transformation signature was temporally stable, indicating an early, persistent tissue state rather than a progressively worsening process. Conclusions. VSD with Dendritic Binary Gating detects a stable pancreatic tissue composition signature on standard CT that is present years before clinical diagnosis, validated across three independent datasets without parameter adjustment. The six sigmoid channels map to biologically meaningful tissue components through a fully transparent interpretability chain. The temporal stability of the signal implies a detection window of 3-7 years, consistent with known PanIN-3 microenvironment transformation timelines. VSD functions as a single-scan screening tool applicable to any abdominal CT performed during the pre-clinical window.

Class imbalance in clinical electrocardiogram (ECG) datasets limits the diagnostic sensitivity of automated arrhythmia classifiers, particularly for rare but clinically significant beat types. We propose a three-stage hybrid generative pipeline that combines a spectral-guided conditional Variational Autoencoder (cVAE), a class-conditional latent Denoising Diffusion Probabilistic Model (DDPM), and a Quantum Latent Refinement (QLR) module built on parameterized quantum circuits to augment minority arrhythmia classes in the MIT-BIH Arrhythmia Database. The QLR module applies a bounded residual correction guided by Maximum Mean Discrepancy minimization to align synthetic latent distributions with real class-specific latent banks. A lightweight 1D MobileNetV2 classifier evaluated over five independent random seeds and four augmentation ratios serves as the downstream benchmark. Our findings establish latent diffusion augmentation as an effective strategy for imbalanced ECG classification and motivate further investigation of quantum-classical hybrid methods in cardiac diagnostics.

As health systems begin deploying autonomous AI agents that make independent clinical decisions and take direct actions within care workflows, ensuring patient safety and care quality requires governance standards that go beyond existing medical device frameworks designed for human-in-the-loop prediction tools. This paper introduces the Healthcare AI Agents Regulatory Framework (HAARF), a comprehensive verification standard for autonomous AI systems in clinical environments, developed collaboratively with 40+ international experts spanning regulatory authorities, clinical organizations, and AI security specialists. HAARF synthesizes requirements from nine major regulatory frameworks (FDA, EU AI Act, Health Canada, UK MHRA, NIST AI RMF, WHO GI-AI4H, ISO/IEC 42001, OWASP AISVS, IMDRF GMLP) into eight core verification categories comprising 279 specific requirements across three risk-based implementation levels. The framework addresses critical gaps in health system readiness for autonomous AI including: (1) progressive autonomy governance with clinical accountability, (2) tool-use security for agents that independently access EHRs, medical devices, and clinical systems, (3) continuous equity monitoring and bias mitigation across diverse patient populations, and (4) clinical decision traceability preserving human oversight authority. We validate HAARFs enforcement capabilities through a scenario-based red-team evaluation comprising six adversarial scenarios executed under baseline (no middleware) and HAARF- guardrailed conditions (N = 50 trials each, Gemini 2.5 Flash primary with Claude Sonnet 4.6 cross-model validation). In baseline conditions, the agent model executes unauthorized tools in 56-60% of adversarial trials. Under the HAARF condition, deterministic middleware enforcement reduces the unauthorized-tool success rate to 0%, with 0% contraindication misses and 0% policy-injection success (95% Wilson CI [0.00, 0.07]). Cross-model validation confirms identical security metrics, supporting HAARFs model-agnostic design. Mapping analysis demonstrates 48-88% coverage of major regulatory frameworks, with per-category FDA alignment ranging from 73% (C5, Agent Registration) to 91% (C3, Cybersecurity; C7, Bias & Equity). Initial validation with healthcare organizations shows a 40-60% reduction in multi-jurisdictional compliance burden and improved clinical safety governance outcomes. HAARF provides health systems with a practical, risk-stratified pathway for safe AI agent deployment--shifting from reactive compliance to proactive quality governance while maintaining rigorous patient safety standards and human-centered care principles.

Dysarthric speech severity assessment typically requires either trained clinicians or supervised machine learning models built from labelled pathological speech data, limiting scalability across languages and clinical settings. We present a training-free method (no supervised severity model is trained; feature directions are estimated from healthy control speech using a pretrained forced aligner) that quantifies dysarthria severity by measuring the degradation of phonological feature subspaces within frozen HuBERT representations. For each speaker, we extract phone-level embeddings via Montreal Forced Aligner, compute d scores along phonological contrast directions (nasality, voicing, stridency, sonorance, manner, and four vowel features) derived exclusively from healthy control speech, and construct a 12-dimensional phonological profile. Evaluating 890 speakers across10corpora, 5 languages for the full MFA pipeline (English, Spanish, Dutch, Mandarin, French) and 3 primary aetiologies (Parkinsons disease, cerebral palsy, amyotrophic lateral sclerosis), we find that all five consonant d features correlate significantly with clinical severity (random-effects meta-analysis rho = -0.50 to -0.56, p < 2 x 10^-4; pooled Spearman rho = -0.47 to -0.55 with bootstrap 95% CIs not crossing zero), with the effect replicating within individual corpora, surviving FDR correction, and remaining robust to leave-one-corpus-out removal and alignment quality controls. Nasality d decreases monotonically from control to severe in 6 of 7 severity-graded corpora. Mann-Whitney U tests confirm that all 12 features distinguish controls from severely dysarthric speakers (p < 0.001).The method requires no dysarthric training data and applies to any language with an existing MFA acoustic model (currently 29 languages) or a model trained from healthy speech alone. It produces clinically interpretable per-feature profiles. We release the full pipeline and phone feature configurations for six languages to support replication and clinical adoption. Author SummaryOne of the authors has lived with ALS for sixteen years. Bernard Muller, who built this entire analytical pipeline using only eye-tracking technology, has experienced the progression of the disease firsthand, including the dysarthric speech that comes with advancing ALS and the tracheostomy that followed. The problem this paper addresses is not abstract to him, and that shapes how the method was designed. We developed a method to measure how well a person with dysarthria can produce distinct speech sounds, without needing any recordings of disordered speech for training. Our approach works by analysing how a widely available AI speech model organises different sound categories -- such as nasal versus oral consonants, or voiced versus voiceless sounds -- and measuring whether those categories become harder to tell apart. We tested this on 890 speakers across 10 datasets in five languages, covering Parkinsons disease, cerebral palsy, and ALS. Because the method only needs healthy speech recordings to set up, it applies to any language with an existing acoustic model, currently covering 29 languages. The resulting profiles show clinicians which specific aspects of speech production are degrading, rather than providing a single opaque severity score. This could support remote monitoring of speech decline in neurodegenerative disease and enable screening in languages and settings where specialist assessment is unavailable.

Background: Current deep learning models in computational pathology, radiology, and digital pathology produce opaque predictions that lack the explainable artificial intelligence (xAI) capabilities required for clinical adoption. Despite achieving radiologist-level performance in tasks from whole-slide image (WSI) classification to mammographic screening, these models function as black boxes: clinicians cannot trace predictions to specific biological features, verify outputs against established morphological criteria, or integrate AI reasoning into precision oncology workflows and tumor board decision-making. Methods: We present Virtual Spectral Decomposition (VSD), a modality-agnostic, interpretable-by-design framework that decomposes medical images into six biologically interpretable tissue composition channels using sigmoid threshold functions - the same mathematical structure as CT windowing. Unlike post-hoc xAI methods (Grad-CAM, SHAP, LIME) applied to black-box deep learning models, VSD channels have pre-defined biological meanings derived from tissue physics, providing inherent explainability without sacrificing quantitative rigor. For whole-slide image (WSI) analysis in digital pathology, we introduce the dendritic tile selection algorithm, a biologically-inspired hierarchical architecture achieving 70-80% computational reduction while preferentially sampling the tumor immune microenvironment. VSD is validated across three cancer types and imaging modalities: pancreatic ductal adenocarcinoma (PDAC) on CT imaging, lung adenocarcinoma (LUAD) on H&E-stained pathology slides using TCGA data, and breast cancer on screening mammography. Composition entropy of the six-channel vector is computed as a visual Biological Entropy Index (vBEI) - an imaging biomarker quantifying the diversity of active biological defense systems. Results: In pancreatic cancer, the fat-to-stroma ratio (a novel CT-derived radiomics biomarker) declines from >5.0 (normal) to <0.5 (advanced PDAC), enabling early detection of desmoplastic invasion before mass formation on standard imaging. In lung cancer, composition entropy from H&E whole-slide images correlates with tumor immune microenvironment markers from RNA-seq (CD3: rho=+0.57, p=0.009; CD8: rho=+0.54, p=0.015; PD-1: rho=+0.54, p=0.013) and predicts overall survival (low entropy immune-desert phenotype: 71% mortality vs 29%, p=0.032; n=20 TCGA-LUAD), providing immune phenotyping for checkpoint immunotherapy patient selection from a $5 H&E slide without molecular assays. In breast cancer, each lesion type produces a characteristic six-channel fingerprint functioning as an interpretable computer-aided diagnosis (CAD) system for quantitative BI-RADS assessment and subtype classification (IDC vs ILC vs DCIS vs IBC). A five-level xAI audit trail provides complete traceability from clinical decision support output to specific biological structures visible on the original images. Conclusion: VSD establishes a unified, interpretable-by-design mathematical framework for explainable tissue composition analysis across imaging modalities and cancer types. Unlike black-box deep learning and post-hoc xAI approaches, VSD provides inherently interpretable, clinically verifiable cancer detection and immune phenotyping from standard clinical imaging at existing costs - without requiring foundation model infrastructure, specialized hardware, or molecular assays. The open-source pipeline (Google Colab, Supplementary Material) enables immediate reproducibility and extension to additional cancer types across the pan-cancer TCGA atlas.

Manuscript preparation is a critical bottleneck in scientific publishing, yet existing AI writing tools require cloud transmission of sensitive content, creating data-confidentiality barriers for clinical researchers. We introduce the Paper Analysis Tool (PAT), a free, multi-agent framework that deploys 31 specialized agents powered by small language models (SLMs) to audit manuscripts across multiple quality dimensions without external data transmission. Applied to three published clinical neurological papers, PAT generated 540 evaluable suggestions. Validation by two expert reviewers (R.B., A.G.) confirmed 391 actionable, high-value revisions (90% agreement), achieving a 72.4% overall usefulness accuracy spanning methodological, statistical, and visual domains. Furthermore, deterministic re-evaluation of 126 agent-suggested rewrite pairs using Phase 0 metrics confirmed text improvement: total word count decreased by 25%, passive voice prevalence dropped sharply from 35% to 5%, average sentence length decreased by 24%, long-sentence fraction fell by 67%, and the Flesch-Kincaid grade improved by 17% . Our validation confirms that systematic, agent-driven pre-submission review drives measurable improvements, successfully converting manuscript optimization from an opaque, manual endeavor into a transparent and rigorous scientific process. Manuscript preparation is a critical bottleneck in scientific publishing, yet existing AI writing tools require cloud transmission of sensitive content, creating data-confidentiality barriers for clinical researchers. We introduce the Paper Analysis Tool (PAT), a free, multi-agent framework that deploys 31 specialized agents powered by small language models (SLMs) to audit manuscripts across multiple quality dimensions without external data transmission. Applied to three published clinical neurological papers, PAT generated 540 evaluable suggestions. Independent validation by two expert reviewers (R.B., A.G.) confirmed 391 actionable, high-value revisions (90% agreement), achieving a 72.4% overall usefulness accuracy spanning methodological, statistical, and visual domains. Furthermore, deterministic re-evaluation of 126 suggested Phase 0 rewrite pairs confirmed text improvement: total word count decreased by 25%, passive voice prevalence dropped sharply from 35% to 5%, average sentence length decreased by 24%, and long-sentence fraction fell by 67%, and the Flesch-Kincaid grade improved modestly. Our validation confirms that systematic, agent-driven pre-submission review drives measurable improvements, successfully converting manuscript optimization from an opaque, manual endeavor into a transparent and rigorous scientific process.

End-stage liver disease (ESLD) is one of the leading causes of death worldwide. Currently, the only curative option for patients with ESLD is liver transplantation. However, the demand for donor livers far exceeds the available supply, partly because many potentially viable livers are discarded following biopsy evaluation. While biopsy is the gold standard for assessing liver histological features related to graft quality and transplant suitability, it often leads to high discard rates due to its susceptibility to sampling errors and limited spatial coverage. Besides, biopsy is invasive, time-consuming, and unavailable in clinical facilities with limited resources. Here, we present an AI-assisted photoacoustic/ultrasound (PA/US) imaging framework for quantitative assessment of human donor liver graft quality and transplant suitablity at the whole-organ scale. With multimodal volumetric PA/US images as the input, our deep-learning (DL) model accurately predicted the risk level of fibrosis and steatosis, which indicate the graft quality and transplant suitability, when comparing with true pathological scores. DL also identified the imaging modes (PAI wavelength and B-mode USI) that correlated the most with prediction accuracy, without relying on ill-posed spectral unmixing. Our method was evaluated in six discarded human donor livers comprising sixty spatially matched regions of interest. Our study will pave the way for a new standard of care in organ graft quality and transplant suitability that is fast, noninvasive, and spatially thorough to prevent unnecessary organ discards in liver transplantation.

Target trial emulation (TTE) enables causal inference from observational data but remains bottlenecked by manual, expert-dependent protocol operationalization. While large language models (LLMs) have advanced clinical knowledge extraction and code generation, their ability to automate end-to-end TTE workflows remains largely unexplored. We present an LLM-driven framework using retrieval-augmented generation to extract the five core TTE design parameters from the Carotid Revascularization and Medical Management for Asymptomatic Carotid Stenosis Trial (CREST-2) protocol and generate executable phenotyping pipelines for real-world EHR data. The performance of the framework was evaluated along two dimensions. First, protocol extraction accuracy was assessed against a gold-standard checklist of trial design components using precision, recall, and F1-score metrics. Second, outcome validity was evaluated through population-level concordance analyses comparing EHR-derived outcomes with published trial endpoints using standardized mean difference, observed-to-expected ratios, confidence interval overlap, and two-proportion z-tests. Further, Human-in-the-loop validation assessed the correctness of extracted clinical logic and phenotype definitions. Together, these evaluations demonstrate a structured approach for assessing LLM-driven protocol-to-pipeline translation for scalable real-world evidence generation.