OBJECTIVE:Low back pain (LBP) is common. Although imaging studies are widely used to evaluate back pain, the link between pain and intervertebral disc degeneration visualized on images is unclear. This study aims to discover novel biomarkers in patients with severe back pain. DESIGN:Levels of 1512 serum proteins from 29 patients with LBP and 11 healthy controls were compared. Machine learning analyses were employed to identify proteins diagnostic of LBP, based on cumulative feature importance across five machine learning models with the highest performance. Proteins with highest feature importance and area under the receiver-operating characteristic curves (AUC) underwent pathway and network analyses. RESULTS:Distinct serum biomarkers differentiating LBP patients from healthy controls were identified. Using Somascan proteomic analysis and machine learning methods, 37 significant biomarkers were highlighted. Notably, IL-19 and biglycan showed the strongest diagnostic potential (AUC 0.91 and 0.89, respectively). Pathway analyses implicate systemic processes in LBP pathology. CONCLUSIONS:IL-19 and biglycan showed promise as serum biomarkers in patients with severe LBP. Given the small sample size of this study, these findings are preliminary and require validation in larger, age- and sex-matched cohorts, with a confirmative, quantitative assay for key markers.

Accurate diagnosis of Alzheimer's disease (AD) requires handling tabular biomarker data, yet such data are often small and incomplete, where deep learning models frequently fail to outperform classical methods. Pretrained large language models (LLMs) offer few-shot generalization, structured reasoning, and interpretable outputs, providing a powerful paradigm shift for clinical prediction. We propose TAP-GPT Tabular Alzheimer's Prediction GPT, a domain-adapted tabular LLM framework built on TableGPT2 and fine-tuned for few-shot AD classification using tabular prompts rather than plain texts. We evaluate TAP-GPT across four ADNI-derived datasets, including QT-PAD biomarkers and region-level structural MRI, amyloid PET, and tau PET for binary AD classification. Across multimodal and unimodal settings, TAP-GPT improves upon its backbone models and outperforms traditional machine learning baselines in the few-shot setting while remaining competitive with state-of-the-art general-purpose LLMs. We show that feature selection mitigates degradation in high-dimensional inputs and that TAP-GPT maintains stable performance under simulated and real-world missingness without imputation. Additionally, TAP-GPT produces structured, modality-aware reasoning aligned with established AD biology and shows greater stability under self-reflection, supporting its use in iterative multi-agent systems. To our knowledge, this is the first systematic application of a tabular-specialized LLM to multimodal biomarker-based AD prediction, demonstrating that such pretrained models can effectively address structured clinical prediction tasks and laying the foundation for tabular LLM-driven multi-agent clinical decision-support systems. The source code is publicly available on GitHub: https://github.com/sophie-kearney/TAP-GPT.

Identifying repurposable therapeutic targets for Alzheimer's disease (AD) remains challenging due to various clinical and biological factors. This study aimed to identify candidate genes for AD therapy. We hypothesize that gene and disease-specific network properties-learnable from these large-scale biomedical knowledge graphs-can inform implicit gene-AD connections and prioritize repurposable AD drug targets. To evaluate the hypothesis, we focused on druggable genes curated from Drug-Gene Interaction Database and Alzheimer's Knowledge Base (AlzKB). We applied scalable random walk methods to Hetionet to learn unbiased gene and disease embeddings, representative of their topological and semantic network properties. The embeddings were then used to compute gene-AD similarity and derive network-based scores for each gene. To validate the scores, using Alzheimer's Disease Sequencing Project (ADSP) data, we constructed AD classifier models with Tree-based pipeline optimizer 2 (TPOT2), an automated machine learning framework. Models were optimized for performance, model complexity, and high aggregate network-based scores. Network-based scores successfully prioritized diverse feature sets-many not previously associated with AD-that are enriched in biologically meaningful body parts such as brain, and pathways including neuronal signaling, potassium channels, and creatine metabolism. The results suggested that knowledge graphs and network-informed embeddings can capture both known and novel insights into AD mechanisms. Additionally, integrating networkbased scores with feature-set-guided TPOT2 offers a scalable and biologically interpretable framework for AD drug repurposing and discovery.

Survival analysis models have evolved significantly with deep learning approaches, yet often lack interpretability and meaningful risk stratification capabilities. We present Interpretable Risk Clustering Intelligence for Survival Analysis (IRIS), a novel framework that addresses the critical task of risk clustering while enhancing both input-level and model-body interpretability. Unlike traditional survival models that perform post-hoc risk clustering, IRIS learns to cluster patients into meaningful risk groups directly from data while providing transparent feature importance estimation through feature contribution functions. We validate IRIS on several benchmark datasets, a real-world Alzheimer's disease dataset, and an electronic health record dataset, showing superior performance in risk clustering and predictive reliability with only a modest decrease in time-toevent prediction accuracy compared to state-of-the-art methods. Our results show that IRIS successfully balances the trade-off between interpretability and prediction performance in riskbased survival analysis, offering clinicians actionable insights for treatment planning and resource allocation.

‘Black box’ deep learning models for medical image interpretation limit clinical trust and analysis of performance degradation. Here we introduce Concept-Level Embeddings for Auditable Radiology (CLEAR), an auditable foundation model based on clinical concepts. Trained on over 0.87 million image–report pairs from 239,391 patients, CLEAR learns a visual representation and projects chest X-rays into a semantically rich space defined by large language model embeddings, making every prediction decomposable into weighted contributions from individual radiological observations. External validation on four large, physician-annotated datasets from the United States, Europe and Asia shows that CLEAR not only achieves state-of-the-art classification performance but also enables applications: auditable zero-shot pathology detection, systematic identification of radiological confounders and the creation of expert-level concept bottleneck models from data-driven concepts. By integrating clinical knowledge directly into its reasoning process, CLEAR offers a framework for robust model auditing, safer deployment and enhanced physician–AI collaboration, advancing towards trustworthy medical AI. CLEAR is an auditable foundation model for chest X-rays that leverages the collective knowledge of the radiological community to offer improved interpretability, performance and new applications.

Single-cell spatial transcriptomics has advanced spatial resolution from several cells per spot to hundreds of transcripts per cell, enabling a more comprehensive understanding of cellular interaction and local tissue organization. However, such high-resolution imaging introduces significant computational challenges, particularly in accurately segmenting cellular boundaries. Existing segmentation methods typically rely on a single modality, such as cellular imaging or transcript profiling, and thus fail to leverage the complementary information between modalities. Here we propose MSCA-Net, a Multi-Scale Convolutional Attention U-Net framework that integrates H&E staining images with selected transcriptomic features to achieve accurate cell boundary extraction. We evaluate MSCA-Net on dorsal root ganglia (DRG) neurons and demonstrate that it consistently outperforms state-of-the-art competing methods. Our study also shows that the reconstructed spatial transcriptomic slice can reproduce the downstream analysis consistent with prior knowledge, providing reliable and valuable insights for biological discovery.

Autoencoders have long been considered a nonlinear extension of Principal Component Analysis (PCA). Prior studies have demonstrated that linear autoencoders (LAEs) can recover the ordered, axis-aligned principal components of PCA by incorporating non-uniform ℓ_2 regularization or by adjusting the loss function. However, these approaches become insufficient in the nonlinear setting, as the remaining variance cannot be properly captured independently of the nonlinear mapping. In this work, we propose a novel autoencoder framework that integrates non-uniform variance regularization with an isometric constraint. This design serves as a natural generalization of PCA, enabling the model to preserve key advantages, such as ordered representations and variance retention, while remaining effective for nonlinear dimensionality reduction tasks.

Accurate in vivo prediction of neuropathology is critical for advancing diagnosis and treatment of Alzheimer’s disease and related dementias (ADRDs). As many individuals with ADRDs have mixed pathologies (β-amyloid, pathologic tau, cerebrovascular disease, vascular brain injury, pathologic TDP-43, hippocampal sclerosis, Lewy bodies), there is interest in determining how accurately we can infer these pathologic changes from clinical data, biofluid assays (e.g., CSF), and neuroimaging. Here we evaluated automated machine learning models trained on data curated by the AD Sequencing Project Phenotype Harmonization Consortium (N=7,894 individuals), to predict 26 autopsy-confirmed neuropathological outcomes. Predictors included in vivo clinical and cognitive composite scores, brain measures from 3D structural MRI and diffusion tensor imaging, image-derived measures of white matter hyperintensities (WMH), and CSF biomarkers. Predictive models were trained using ensemble learning with stratified cross-validation. We assessed performance using Spearman’s rank correlation and Matthews correlation coefficient, to accommodate co-occurring pathologic changes. The added value of neuroimaging and CSF versus clinical features alone was quantified. Braak stage was among the most consistently predicted outcomes. CSF biomarkers best predicted β-amyloid and tau pathology, but diffusion MRI metrics best captured vascular brain injury and white matter injury, and outperformed clinical and cognitive measures and anatomical MRI in predicting Lewy body disease. Anatomical measures from structural MRI outperformed standard clinical assessments in assessing neurodegeneration and hippocampal sclerosis, and WMH complemented cognitive measures in predicting TDP-43 pathology. These results establish a baseline for comparing modalities for inferring neuropathology.

In recent years, rapid technological advances have made artificial intelligence a global trend, significantly advancing the fields of medicine and healthcare. Meanwhile, concerns about information security and medical data privacy have also increased. Although many studies have collected data from hospitals to train centralized models, data sharing between medical facilities is often restricted and protected by law. Federated learning (FL) emerges as a promising solution, enabling model training across different medical facilities without sharing raw data, thereby safeguarding patient privacy and ensuring data security. Our study reviews key works on FL in medical image analysis (MIA) from 2020 to 2025. We systematically analyze major challenges, including data heterogeneity, non-independent and identically distributed data, missing labels, communication costs, and privacy risks during model updates. Additionally, this review covers emerging approaches such as semi-supervised learning, unsupervised learning, domain shift learning, advanced deep learning architectures, and personalized models within the context of FL. We aim to provide a comprehensive overview of the current research landscape, methods for addressing these challenges, and future directions for FL in MIA.

Predicting the trajectory of clinical decline in aging individuals is a pressing challenge, especially for people with mild cognitive impairment, Alzheimer’s disease, Parkinson’s disease, or vascular dementia. Accurate predictions can guide treatment decisions, identify risk factors, and optimize clinical trials. In this study, we compared two deep learning approaches for forecasting changes, over a 2-year interval, in the Clinical Dementia Rating scale ‘sum of boxes’ score (sobCDR), as a continuous outcome (regression). This is a key metric in dementia research and clinical trials, and scores range from 0 (no impairment) to 18 (severe impairment). To predict decline, we trained a hybrid convolutional neural network (CNN) that integrates 3D T1-weighted brain MRI scans with tabular clinical and demographic features (including age, sex, body mass index (BMI), and baseline sobCDR). We benchmarked its performance against AutoGluon, an automated multimodal machine learning framework that selects an appropriate neural network architecture (an ‘autoML’ approach). We evaluated the models using data from 2,319 unique participants drawn from three independent cohorts—ADNI, OASIS-3, and NACC. For each participant, we used one T1-weighted brain MRI scan along with corresponding clinical and demographic information. Our results demonstrate the importance of combining image and tabular data in predictive modeling for this clinical application. Deep learning algorithms can fuse information from image-based brain signatures and tabular clinical data, with potential for personalized prognostics in aging and dementia. Rather than concluding that multimodal fusion uniformly improves performance, our results show that deep learning applied to volumetric MRI data may struggle to add predictive value, particularly when clinical covariates explain substantial variance and provide a strong baseline. In other conditions and tasks, it may help to have a hybrid system that can learn from both data types, and their relative value may be different. Conversely, AutoML-based multimodal fusion provides a robust baseline when tabular data already provide strong predictive value for the task. These insights clarify how different multimodal strategies could be selected in clinical prognostic applications.

While biological and pharmaceutical knowledge networks have significantly propelled drug repurposing efforts, reliance solely on these networks is insufficient for accurately addressing genetic and phenotypic variance. This limitation highlights the need for an integrative approach that leverages context-specific data to enhance the precision of drug repurposing. We introduce a network-based integrative drug scoring approach that synergistically incorporates data-driven and knowledge-driven networks without requiring their direct integration. We developed a synergistic label propagation algorithm that facilitates information transfer from data-driven to knowledge-driven networks. To enable context-specific drug repurposing, we constructed a data-driven disease-disease association network utilizing European-specific genetic information and a knowledge-driven drug-target protein association network. In a proof-of-concept study, drug scoring was applied to identify candidate drugs for rheumatoid arthritis, asthma, and multiple sclerosis. Compared with a representative direct-integration benchmark, the proposed method achieved an average AUC of 0.701, corresponding to a 9.71

BACKGROUND:Single-cell RNA sequencing has emerged as a powerful approach to reveal cellular heterogeneity within biological systems. With the continuous advancement of high-throughput sequencing technologies, studies are generating vast amounts of complex data, posing a significant challenge for researchers in effective data processing and analysis. RESULTS:To address this issue, we developed SCSEQ, an interactive web-based bioinformatics analysis platform. This platform enables even users without programming expertise to conveniently process and analyze sequencing data. SCSEQ provides a comprehensive workflow encompassing data preprocessing, normalization, clustering, dimension reduction, differential expression analysis, cell type identification, and downstream analyses. The downstream analysis tasks include gene enrichment analysis, transcription factor analysis, cell-cell communication analysis, copy number variation detection, trajectory inference, and pan-cancer analysis. SCSEQ facilitates information transfer between different workflows, accepts various input formats, and generates graphical and tabular outputs. As a user-friendly platform, we enhance user experience through detailed parameter settings and dynamic interactions. This enables users to precisely regulate research processes and customize result figures. Additionally, we provide comprehensive user manuals to assist with parameter configuration and workflow execution. CONCLUSIONS:SCSEQ provides an intuitive and convenient solution for single-cell transcriptome sequencing data analysis. Our platform has successfully completed full-process analyses on real-world data with reliable results, demonstrating its applicability in practical scenarios. The platform is available at https://scseq.com.cn/.

With the advent of high-throughput techniques, multi-omics data and various clinical outcomes have been collected for a range of diseases. Multi-omics data play a crucial role in uncovering complex biological processes, yet simultaneous representation learning of such high-dimensional, heterogeneous multi-modality data along with clinical outcomes remains limited. To address this gap, we propose a supervised knowledge-guided Bayesian factor model for integrative analysis of multi-omics and clinical outcome data. The proposed method simultaneously extracts an informative low-dimensional representation and predicts one or more clinical outcomes of interest. The two-level adaptive shrinkage in the novel hierarchical priors allows for the identification of both active modalities and features, resulting in a biologically meaningful structural identification of the high-dimensional data. Moreover, the method is robust to noisy edges in biological graphs that do not align with ground truth. Finally, the proposed method can handle different data types including both continuous and categorical data. Extensive simulation studies and real data analyses of Alzheimer's disease (AD) data demonstrate the advantages of the proposed approach over existing methods. Notably, our analysis of multi-omics and imaging phenotype data from ADNI provides meaningful insights into the underlying biological mechanisms of AD.

Cross-entropy (CE) is the default training loss for supervised classification, but its sample efficiency is limited when labels are scarce. Existing remedies primarily act on the data side, via augmentation, synthesis, or transfer from pretrained models; the training objective itself is rarely revisited. We revisit it here. Drawing on the classical observation that generative classifiers reach their asymptotic error with fewer samples than discriminative ones, we propose Generative Cross-Entropy (GenCE), a drop-in replacement for CE that introduces a generative learning principle into a standard discriminative network without altering the architecture or fitting a separate density model. GenCE follows from a Bayesian rewrite of the class-conditional likelihood and, in the mini-batch approximation, reduces to normalizing each sample's softmax score against the model's predictions on the batch, coupling the training signal across examples sharing a class. We extend the proper-scoring-rule framework to such non-local losses and prove that GenCE is strictly proper under a mild completeness condition: its population risk is uniquely minimized at the true posterior. Across three datasets, on two architectures and in both balanced small-data and class-imbalanced regimes, GenCE outperforms CE and other widely used losses, while also producing better-calibrated probabilities and stronger out-of-distribution detection.

Alzheimer's disease (AD) patients suffer from consequential diagnostic delay due to the lack of accessible biomarkers. They also show different responses to treatments due to disease heterogeneity and progression. Here, we developed a novel framework to identify disease progression and subtypes by using geometric brain signatures derived from multiple neuroimaging modalities, including [ 18 F]-Florbetapir (AV45) Positron Emission Tomography (PET), [ 18 F]-Fludeoxyglucose (FDG) PET, and structural Magnetic Resonance Imaging (MRI). These signatures were derived by decomposing corresponding maps of amyloid-beta levels, metabolic activity, and cortical thickness in terms of the fundamental, resonant modes--eigenmodes--of cortical geometry, each tied to a specific spatial resolution scale. Our results showed that geometric eigenmode-based features identified trajectories of disease progression, quantified as pseudotime, in distinct subtypes. The disease progression trajectories and subtypes are identified with high stability and are highly related to biological and cognitive measures. These performances are superior to those obtained using conventional localised features and remain robust across datasets, indicating that geometric signatures of brain structure and function can be used to uncover new markers of AD diagnosis and prognosis that are missed by conventional localisation approaches.

Large Language Models (LLMs) have shown significant promise for clinical applications, yet their application to triage remains underexplored. In this study, we systematically investigate the capabilities of LLMs in emergency department triage through two key dimensions: (1) robustness to distribution shifts and missing data, and (2) intersectional biases across sex and race. We assess multiple LLM-based approaches, ranging from continued pre-training to in-context learning, as well as conventional machine learning (ML) approaches. First, we demonstrate that LLMs exhibit superior robustness compared to traditional ML, which is promising due to their ability to provide explanatory rationales. Second, we show that the most effective LLM-based methods are those that select similar examples from prior patient cases, whereas reasoning capabilities in LLMs offer little benefit for triage. Lastly, we identify critical gaps in LLM preferences that emerge at the intersections of sex and race. LLMs exhibit sex-based differences, and they are more pronounced in certain racial groups, suggesting that LLMs encode preferences that emerge in specific clinical contexts and combinations of characteristics. We perform this audit through counterfactual analysis, providing a systematic way to identify such biases before real-world integration.

Background Recent advancements in single-cell omics technologies have enabled detailed characterization of cellular processes. However, coassay sequencing technologies remain limited, resulting in unpaired single-cell omics datasets with differing feature dimensions.Finding We present GROTIA (Graph-Regularized Optimal Transport Framework for Diagonal Single-Cell Integrative Analysis), a computational method to align multi-omics datasets without requiring any prior correspondence information. GROTIA achieves global alignment through optimal transport while preserving local relationships via graph regularization. Additionally, our approach provides interpretability by deriving domain-specific feature importance from partial derivatives, highlighting key biological markers. Moreover, the transport plan between modalities can be leveraged for post-integration clustering, enabling a data-driven approach to discover novel cell subpopulations.Conclusions We demonstrate GROTIA's superior performance on four simulated and four real-world datasets, surpassing state-of-the-art unsupervised alignment methods and confirming the biological significance of the top features identified in each domain.

Chronic psychological stress has been implicated as a risk factor for Alzheimer’s disease (AD), potentially through cortisol-mediated acceleration of disease progression. However, the molecular pathways underlying this relationship remain poorly understood. Epigenetic regulation of the glucocorticoid and mineralocorticoid receptor genes (NR3C1 and NR3C2), which encode receptors for cortisol, may play an important role, but has not been examined in relation to AD progression. Therefore, this study investigated associations between DNA methylation of NR3C1/NR3C2 and AD-related phenotypes, including cognition, brain amyloid-β (Aβ) burden, and regional brain volumes. These associations were examined in two independent cohorts of cognitively unimpaired individuals with accumulating brain Aβ (n = 89–298 across outcomes) using linear regression and meta-analyses. The study also explored whether DNA methylation within NR3C1 and NR3C2 interacted with depression symptoms to influence relationships with AD-related phenotypes. While only nominal associations were observed in direct analyses, stronger associations emerged in interaction with depressive symptoms. Interaction analyses showed that relationships between DNA methylation and AD-related phenotypes (cognition, hippocampal volume and ventricular expansion) differed depending on the presence of depression symptoms. Consistent patterns across cohorts were observed, with associations primarily evident among individuals with clinically relevant depressive symptoms. One site (NR3C1 cg24052866) was associated with cognitive decline, one (NR3C1 cg08845721) with cross-sectional hippocampal volume, and eight (NR3C1 cg21979215, cg16594263; NR3C2 cg27460943, cg17253842, cg04867484, cg10993059, cg25672354, cg27234800) with ventricular expansion. These exploratory findings suggest epigenetic variation within cortisol receptor genes may influence AD-related neurodegeneration in a depression-dependent manner.

Gaps remain in our understanding of how large language models (LLMs) acquire knowledge during pre-training. We posit that auxiliary views, reformulations of knowledge, are causally helpful for learning. We design controlled experiments to isolate this. First, we confirm that repetition is necessary for acquisition and clarify that paraphrasing helps only at smaller batch sizes. Second, holding the token budget fixed, allocating tokens from document repetition to auxiliary views improves learning, counterintuitively, even for factual recall. Third, the effectiveness of auxiliary views is not contingent on the strength of the teacher model that generates them. Fourth, we identify forms of knowledge, contextual and foundational, that aid learning in the presence of prior knowledge gaps. Finally, we examine how these effects manifest mechanistically via layer-wise biases and compression. Together, our findings suggest that auxiliary representations of knowledge, which arise naturally in large pre-training corpora, are a key factor in the success of pre-training and offer a plausible explanation for why data diversity matters.