Hyperspectral-multispectral image fusion (HMIF) has emerged as a long-standing and widely studied research topic, providing an effective pathway to address the intrinsic trade-off between spatial and spectral resolutions in single-sensor imaging systems. Yet, most existing approaches are confined to the image or feature domain, where the fusion strategies in such a pixel-aware space fail to model spectral variability and avoid high-dimensional redundant computations. To overcome these difficulties, we propose performing the fusion process in the state space, a basis space that describes the variation induced by cross-resolution using a learnable controlled matrix, yielding a novel multimodal fusion framework, termed Im2State. More specifically, Im2State comprises a series of cross-state fusion blocks, in which modality-specific state-transition matrices are cross-coupled and jointly optimized. This design enables the unified modeling of multimodal information injection within the state space, allowing for the fine-grained regulation of modality contributions and the dynamic correction of fusion-induced errors throughout the reconstruction process. Furthermore, Im2State incorporates a multi-stage cross-modal alignment strategy with composite consistency constraints to mitigate structural discrepancies caused by resolution mismatch and modality heterogeneity, thereby enhancing the consistency and stability of state evolution. Extensive experiments on well-known six benchmark datasets demonstrate that Im2State consistently outperforms state-of-the-art methods in both quantitative performance and visual quality, validating the proposed cross-state fusion strategy as a principled and effective paradigm for high-fidelity HMIF.

ZNRF3 and RNF43 are closely related transmembrane E3 ubiquitin ligases with significant roles in development and cancer. Conventionally, their biological functions have been associated with regulating WNT signaling receptor ubiquitination and degradation. However, our proteogenomic studies have revealed EGFR as the protein most negatively correlated with ZNRF3/RNF43 mRNA levels in multiple human cancers. Through biochemical investigations, we demonstrate that ZNRF3/RNF43 interact with EGFR via their extracellular domains, leading to EGFR ubiquitination and subsequent degradation facilitated by the E3 ligase RING domain. Overexpression of ZNRF3 reduces EGFR levels and suppresses cancer cell growth in vitro and in vivo, whereas knockout of ZNRF3/RNF43 stimulates cell growth and tumorigenesis through upregulated EGFR signaling. Together, these data suggest ZNRF3 and RNF43 as novel E3 ubiquitin ligases of EGFR and establish the inactivation of ZNRF3/RNF43 as a driver of increased EGFR signaling, ultimately promoting cancer progression. This discovery establishes a connection between two fundamental signaling pathways, EGFR and WNT, at the level of cytoplasmic membrane receptors, uncovering a novel mechanism underlying the frequent co-activation of EGFR and WNT signaling in development and cancer.

Models based on Convolutional Neural Networks (CNNs) and Transformers have made significant advancements in the joint classification of hyperspectral image (HSI) and Light Detection and Ranging (LiDAR) data. However, these models also exhibit inherent limitations. CNNs, constrained by their network architecture, struggle to effectively capture long-range feature dependencies. On the other hand, while Transformers, with the self-attention mechanism, are adept at modeling global features, their scalability is hindered by the quadratic computational complexity. Fortunately, the recently emerged Mamba, a State Space Model (SSM), offers high computational efficiency while retaining the modeling capabilities of Transformers. In this paper, we present a preliminary attempt to apply the Mamba architecture to the joint classification task of HSI and LiDAR data. We utilize a dual-branch Siamese Visual State Space Block to extract distinctive features from the two heterogeneous modalities. The core module, 2D Selective Scan (SS2D), traverses paths in four different directions to generate multiple image patch sequences, with each sequence processed by individual S6 blocks. These sequences are then integrated and merged to capture the global features of the image. Considering the importance of inter-modal complementarity in improving classification performance, we introduce two lightweight Mamba feature extraction and fusion modules. The Cross-Mamba Feature Extraction Block is designed to enhance cross-modal information flow through features interaction. To better select relevant information from each modality, the Modality-Guided Mamba Fusion Block employs an attention mechanism to highlight pertinent information, further enhancing the model's discriminative ability. The aim is to reduce the number of training parameters while maintaining competitive classification performances. E-Mamba represents a pioneering effort in applying the Mamba architecture to the joint classification of HSI and LiDAR data, providing a foundation for future networks in multi-source remote sensing (RS) image classification based on Mamba. We demonstrate the effectiveness of E-Mamba through extensive experiments on three real-world datasets, achieving overall classification accuracies of 93.81%, 99.06%, and 91.47% on the Houston2013, Trento, and MUUFL datasets, respectively, surpassing a range of latest CNN-based and Transformer-based models. The code will be released at https://github.com/zhangyiyan001/E-Mamba.

Given escalating water scarcity and rapid urbanization, characterizing the long-term evolution of surface water environments across the Beijing-Tianjin-Hebei (BTH) region is vital for coordinated regional development. Using Landsat imagery spanning 1984–2024, this study quantified Secchi disk depth (Zsd) dynamics in 1105 water bodies across the BTH region and identified the dominant drivers through correlation analysis and generalized linear models (GLMs). Results show that water bodies in the BTH region are predominantly small (1.33 ± 10.32 km2) and characterized by a generally low baseline water clarity (mean Zsd = 0.62 ± 0.22 m). Nevertheless, a pervasive long-term clearing trend was detected: 60.0% of water bodies (N = 663) exhibited significant increases in water clarity over the past four decades, whereas only 4.2% (N = 46) showed significant declines (P < 0.05). GLM analysis revealed that water area expansion was the primary driver of Zsdvariability, consistently explaining more than 45% of the long-term variations. This expansion was closely linked to precipitation variability in Hebei Province, while in Beijing and Tianjin it was primarily driven by the South-to-North Water Diversion Project. Land-cover transitions acted as secondary drivers, accounting for 27.2% of the observed variance, whereby reductions in barren land and increases in impervious surfaces together with strengthened environmental governance, effectively mitigated sediment loading. In contrast, intensive cropland in certain regions, particularly Hebei, may still exert a localized influence, where agricultural runoff can partially limit improvements in water clarity. Overall, these findings suggest that strategic water diversion and ecological restoration have been associated with improvements in regional surface water quality, while localized agricultural pressures require more targeted management interventions. This study provides new insights into how both natural and anthropogenic drivers interact to shape long-term water clarity dynamics in this highly water-stressed region and offers a robust scientific basis for informed water resource and land-cover management, and sustainable regional development.

Abstract Over 90% of human genes undergo alternative splicing, generating numerous transcripts or isoforms with distinct functions for each gene. This highly regulated process is often disrupted in cancer, leading to the production of harmful protein isoforms that contribute to tumor growth, survival, metastasis, and immune evasion. Accurately identifying these aberrant isoforms is essential for understanding cancer biology and developing targeted therapies. Although RNA sequencing has advanced our understanding of alternative splicing in cancer, the study of protein isoforms at the proteomic level is crucial as mRNA expression and protein abundance are only moderately correlated. Advancements in mass spectrometry (MS)-based shotgun proteomics have enabled unbiased identification and quantification of more than 10,000 protein coding genes from biological sample, as demonstrated in the proteomic datasets from the Clinical Proteomic Tumor Analysis Consortium (CPTAC). However, typical shotgun proteomics analysis is done at the gene level and is unable to quantify protein isoforms. To better use protein isoform information from CPTAC global proteomics data set, we published a tripartite graph modeling approach that groups peptides based on their mapping relationships to protein isoforms. In this study, we utilized our recent algorithm to analyze global proteomics data from ten tumor cohorts from CPTAC to generate protein isoform or protein isoform group level quantification matrices. These matrices were integrated with other omics, including somatic mutation, copy-number variation, methylation, gene expression, and gene level protein abundance, and clinical data from the harmonized dataset of the CPTAC pan-cancer project. After constructing this dataset, at first, protein isoforms were compared between tumor and adjacent normal tissues, linked to phenotype, and correlated with other omics; Then, we investigated the effects of transcription factors and mutated splicing factors on protein isoforms; Moreover, we performed eQTL analysis using protein isoform abundance and isoform level mRNA expression. Through this approach, we identified numerous alternative splicing events that drive human tumors. Overall, our study provides a comprehensive characterization of protein isoforms in human cancers, offering deep insights into alternative splicing and its potential for advancing cancer biology and targeted therapy development. Citation Format: Yongchao Dou, Lindsey Olsen, Bing Zhang. Pan-cancer characterization of protein isoforms uncovers driving alternative splicing events at protein level [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 7687.

566 Background: Endocrine therapy (ET) resistance (ETR) remains a primary challenge in ER+ breast cancer. Analyzing pretreatment tumor transcriptomes across trials with early response endpoints can reveal shared and specific ETR signatures. This study utilizes baseline RNA data from the Phase III ALTERNATE trial (Alliance A011106, NCT01953588; Anastrozole [A], Fulvestrant [F], or AF) and the ACOSOG Z1031B trial (NCT00824941) to identify predictors of early Ki67 response in postmenopausal ER+/HER2– patients. Methods: ETR was defined as week-4 on-treatment Ki67 >10%. Baseline gene expression from ALTERNATE was analyzed to identify differentially expressed (DE) genes (Wilcoxon test, P<0.05) and Hallmark pathways associated with ETR, both across and within individual treatment arms. Feature selection was performed using mixOmics. A Pan-Endocrine Therapy Signature (PETS) was developed by uniting DE genes identified across all three ALTERNATE arms and Z1031B. All statistical analyses were conducted in R (P<0.05). Results: Overall ETR rate in the ALTERNATE RNA-seq cohort (n=733) was 26%. ETR was associated with high Risk of Recurrence (ROR), Oncotype RS, and Mammaprint scores (calculated from RNA-seq data in research setting). In luminal tumors (n=649), ETR was linked to chr 3q13.33, 8q24.13, and 20q13.12 cytoband upregulation and 17q21, 18q23, 3p21.1, and 10q24.32 cytoband downregulation. ETR tumors showed T-cell, E2F target, and interferon-γ enrichment; sensitive tumors favored early estrogen response and muscle differentiation. At individual gene level, high MYBL2 , PIF1 , TROAP and with low HJURP predicted ETR across all samples (AUC>0.70). A deep learning model using all protein-coding genes achieved AUC 0.82 (training) and 0.79 (test) in predicting ETR. Cross-trial integration identified ETR-associated PETS, enriched for genomic instability. PETS performed comparably to established signatures and strongly correlated with MYBL2 signature (r=0.93). Top ETR predictors were MYBL2 , AURKB , and EME1 for Arm A and IL4I1 , TNFAIP6 , and ANLN for Arm AF. AF-resistant tumors were enriched for systemic lupus and RIG-I–like receptor signaling; sensitive tumors favored PI3K–AKT, EGFR TKI resistance, AMPK, and insulin signaling. Conclusions: Baseline transcriptomics identify shared and therapy-specific ETR markers. The 15-gene PETS defines a convergent resistance signature, performing similar to established signatures in predicting ETR and correlating with MYBL2. Enrichment of cell-cycle and immune pathways in resistant tumors may suggest patient stratification approach for alternative or combinatorial strategies to overcome early ETR in ER+ breast cancer. Acknowledgement: https://acknowledgments.alliancefound.org. Support: U10CA180821, U10CA180882, U24CA1. Clinical trial information: NCT01953588 .

Abstract Phosphorylation is a central regulator of protein function and oncogenic signaling, and advances in mass spectrometry now enable unbiased, proteome-wide identification of cancer-associated phosphosites. However, the functional relevance of most sites remains poorly understood and scattered across the literature.To address this gap, we developed PTMax, an AI-enabled resource that integrates comprehensive literature mining with systematic multi-omics data to advance functional interpretation of phosphorylation in cancer.To standardize phosphosite information reported across published studies, we enhanced our literature-mining pipeline to efficiently extract site-level evidence and associated functional information from full-text articles and pathway figures. Evidence aggregated from these sources was used to generate functional summaries, which were evaluated through both automated and manual quality assessments. PTMax additionally incorporates dozens of mass spectrometry-based phosphoproteomics datasets and multi-omics data from CPTAC cancer cohorts, including RNA, protein, phosphosite abundance, and phenotype associations. For each phosphosite, we computed two evidence scores that quantify literature-derived information richness and data-driven support, respectively. We also constructed signature sets that group phosphosites by cancer hallmarks, co-mentioned genes or diseases, and pathway figure associations, and generated a co-regulated phosphorylation network to facilitate pathway- and network-level interpretation.PTMax currently contains more than 40,000 literature-derived phosphosites extracted from over 500,000 sentences and 1,400 pathway figures, capturing ∼70% of low-throughput-validated and ∼80% of regulatory-annotated phosphosites in PhosphoSitePlus. Notably, over 30,000 sites lack prior regulatory evidence, underscoring the value of AI-driven literature mining. Integration with multi-omics resources adds ∼200,000 unique phosphosites, including 65,000 sites with quantitative associations and 26,000 linked to cancer phenotypes. The PTMax interface enables users to query individual genes or phosphosites and retrieve comprehensive, context-rich information together with user-friendly visualizations. In addition, pathway and network-based analysis modules help translate phosphosite lists into functional and signaling insights. In summary, PTMax unifies literature and figure mining with large-scale experimental datasets to deliver a comprehensive, multi-dimensional resource that advances the functional study of phosphorylation in cancer. Citation Format: Yanling Sun, Sara S. Savage, John M. Elizarraras, Eric Jaehnig, Bing Zhang, . PTMax: An AI-enabled platform integrating literature mining and multi-omics for functional interpretation of phosphorylation in cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2703.

Isobaric mass tags, such as isobaric tags for relative and absolute quantitation (iTRAQ) and tandem mass tag (TMT), are widely utilized for peptide and protein quantification in multiplex quantitative proteomics. We present TMT-Integrator, a bioinformatics tool for processing quantitation results from TMT and iTRAQ experiments, offering integrative reports at the gene, protein, peptide, and post-translational modification site levels. We demonstrate the versatility of TMT-Integrator using five publicly available TMT datasets: clear cell renal cell carcinoma (ccRCC) whole proteome and phosphoproteome datasets from the Clinical Proteomic Tumor Analysis Consortium, an E. coli dataset with 13 spike-in proteins, and two human cell lysate datasets showcasing the latest advances with the Thermo Orbitrap Astral mass spectrometer and TMTpro 35-plex reagents. Integrated into the widely used FragPipe computational platform ( https://fragpipe.nesvilab.org/ ), TMT-Integrator is a core component of TMT and iTRAQ data analysis workflows. We evaluated the performance of FragPipe coupled with TMT-Integrator analysis pipeline against MaxQuant and Proteome Discoverer with multiple benchmarks, facilitated by the bioinformatics tool OmicsEV. Our results show that FragPipe coupled with TMT-Integrator quantifies more proteins in the E. coli and ccRCC whole proteome datasets, quantifies more phosphorylated sites in the ccRCC phosphoproteome dataset, and overall delivers more robust quantification performance compared to other tools.

The rapid advancement of deep neural networks (DNNs) has substantially progressed image-to-image translation, yielding numerous sophisticated methods. However, most existing methods face not only the inherent pixel-level spatial misalignment resulting from divergent imaging perspectives, but also the local geometric distortion and structural incoherence stemming from inadequate cross-modal feature alignment. To address this issue, we propose CycleMamba, a cycle-consistent learning-based aerial visible-to-infrared image translation framework, which enforces geometric constraints and semantic space alignment through globally-aware bidirectional transformation, thereby alleviating pixel-level misalignment and structural distortion. Specifically, inspired by the selective structured state-space model (SSM, Mamba), a bidirectional cross-modal translation network based on multigranularity U-shaped translators (MGUTs) is constructed, which integrates Mamba's long-range modeling with convolutional neural network (CNN) local feature extraction strengths. Regarding the stability of cyclic consistency learning, a dual-stage progressive training mechanism is developed for visible-infrared-visible translation. Additionally, to enhance the alignment of cross-modal features and structural preservation, the cycle consistency constraints that collaborate with structural similarity (SSIM) and semantic consistency losses are given to reduce spatial and semantic misalignment, facilitating fidelity. Comparative experiments with state-of-the-art methods are conducted on three public datasets. The experimental results demonstrate that CycleMamba achieves superior translation performance. Extensive ablation studies further evaluate the effectiveness of the proposed method. The code will be available at https://github.com/xzhichaox/CycleMamba

Tiny object detection (TOD) in remote sensing imagery remains challenging because foreground signals are extremely weak in deep feature hierarchies and are easily overwhelmed by high-response background interference. To mitigate this observed foreground-background signal modulation imbalance (FBSMI) difficulty, we propose a signal modulation network (SMN) for remote-sensing TOD. SMN comprises two complementary components. First, an adaptive Wiener filter modulator (AWFM) is inserted after backbone stages to suppress background-dominated noise while preserving weak target-related responses at multiple resolutions. Second, we introduce the novel denoising diffusion transformer (DDT), a featurespace conditional diffusion module that operates on detector feature tensors rather than image pixels. DDT generates multiple diffusion-guided semantic feature variants from high-level fused features and expands the local representation space around weak tiny object evidence. Extensive experiments on AI-TOD, SODA-A, DOTAv2.0, and DIOR-R demonstrate that SMN not only effectively mitigates the FBSMI problem, but also improves detection accuracy, particularly for very tiny and tiny objects, compared with state-of-the-art methods.

With the rapid increase in hyperspectral images (HSIs), hyperspectral object-level detection (HOD) has become an important task. However, existing methods often neglect the material similarity between objects and the background, which increases the difficulty of object-background discrimination. Meanwhile, they have not sufficiently exploited the intrinsic advantage of HSIs, namely their continuous full-spectrum representation, which jointly captures global spectral context, local spectral variations, and dependencies across bands. To address these limitations, this paper proposes the SPG-OD method by integrating spectral prior information. We introduce two key modules: the Grouped Local Spectral Enhancement Module (GLSEM), which enhances local spectral variations to improve object-background discrimination when objects and background share similar materials, and the Spectral Objectness Prior Module (SOPM), which uses prior spectral curves to guide detection. Experimental results on three large-scale HOD datasets show that SPG-OD achieves competitive overall performance and demonstrates the effectiveness of incorporating spectral prior information for hyperspectral object detection.

Provides society information that may include news, reviews or technical notes that should be of interest to practitioners and researchers.

The bidirectional reflectance factor (BRF) characterizes the directional radiative properties of terrestrial surfaces. However, existing three-dimensional (3D) radiative transfer models require complex scene construction and computationally intensive radiative transfer solvers, limiting efficient generation of multi-angle hyperspectral reflectance imagery. 3D Gaussian Splatting (3DGS) offers an efficient framework for neural scene representation and novel view synthesis, but its low-order spherical harmonics representation is insufficient for complex directional reflectance, while the high dimensionality and inter-band quality differences of hyperspectral data introduce additional challenges. To address these challenges, we propose BRF-GS, a 3DGS-based framework for BRF modeling and hyperspectral reflectance image generation. BRF-GS introduces a hybrid BRDF-driven kernel to represent complex directional reflectance, selects geometry-reliable spectral bands for robust 3D scene initialization, and adopts a two-stage training strategy that decouples geometry optimization from spectral modeling. We further construct the AIR-BRF dataset, a multi-angle hyperspectral directional reflectance dataset comprising three scenes with diverse natural and artificial targets. Experiments demonstrate that BRF-GS achieves superior spatial and spectral fidelity and accurately reproduces characteristic view-dependent BRF responses. The proposed framework provides an efficient data-driven approach for BRF modeling and multi-angle hyperspectral reflectance image generation in remote sensing scenes.

Hyperspectral images generally suffer from low spatial resolution, limiting their utility in fine-scale applications. To address the limitations of existing unsupervised fusion methods in adapting to spatial heterogeneity and scale effects that often lead to blurred details and reduced target discrimination, we propose an unsupervised adaptive scale-aware detail feature extraction network (UASNet), which introduces a structure-adaptive mechanism and degradation-aware modeling to effectively harmonize the representation consistency of multiscale ground objects without paired training samples. The network consists of three stages: the prior information mining stage, the spectral channel mapping stage, and the detail feature fusion stage. Specifically, an adaptive scale-aware convolution is embedded within a reversible detail extraction module to capture features of objects with varying scales and geometries, thereby preserving fine textures and structural integrity. Furthermore, driven by the requirements of real-world application scenarios, spectral and spatial branches are constructed to learn the respective degradation priors, which are integrated into the loss function to realize a fully unsupervised fusion framework. Extensive experiments on both simulated and real datasets demonstrate the superior performance of the proposed method compared with that of state-of-the-art approaches. Moreover, even without ground truth data, the downstream classification results indirectly validate that UASNet exhibits strong potential for real-world applications.

Due to the high cost of data acquisition and annotation, it is challenging to obtain large-scale training data for hyperspectral remote sensing images (HRSIs), making it difficult to train object detectors directly. In contrast, visible-light remote sensing images (VLRSIs) are abundant and well-annotated. Hence, cross-domain object detection from VLRSI to HRSI (VLRSI2HRSI) provides a feasible solution for HRSI detection. However, existing cross-domain detection methods are developed for natural images. They neglect the shared spatial characteristics in the remote sensing scenario and fail to model the spectral properties of HRSIs. As a result, directly applying them to VLRSI2HRSI leads to significant performance degradation. To address this issue, we propose a Visible-light to Hyperspectral Cross-Domain Detection Network (VHCDN). We first observe that reconstructed residual representations in both VLRSI and HRSI suppress background and emphasize foreground objects. Based on this observation, we introduce a Residual Extracting Module (REM) to extract domain-shared spatial features. Furthermore, since objects of the same semantic class may consist of different materials and exhibit spectral variations under varying illumination, we design a Spectral Saliency Head (SSH) that models spectral uncertainty. By introducing one HRSI into the source-domain VLRSI datasets to guide the generation of spectral features, the SSH is trained with a spectral heteroscedastic loss and a spectral uncertainty loss, enabling the uncertainty learned in the labeled source domain to be stably transferred to the unlabeled target domain. Experiments on three cross-domain settings against seven representative methods demonstrate that VHCDN achieves state-of-the-art performance, while ablation studies verify the effectiveness of each proposed module.

Timely and accurate estimation of regional winter wheat yield is of great significance for safeguarding food security and promoting sustainable agricultural development. In recent years, deep learning has been widely applied in crop yield estimation due to its powerful capability in mining complex relationships. However, the irregular shapes of administrative regions pose challenges for integrating spatial data such as remote sensing into deep learning models. To address this issue, this study employed mean-based aggregation and histogram-based dimensionality reduction techniques to preprocess spatial data, including remote sensing and meteorological data, thereby generating sample sets suitable for deep learning models. This study identified the phenomenon of feature conflict when processing heterogeneous features in conventional Long Short-Term Memory (LSTM) models and proposed a TB-LSTM (Two-Branch LSTM) model to mitigate such conflicts. The impact of different input feature combinations on estimation accuracy was analyzed, and the model’s capability for early yield prediction was further evaluated. The results show that: (1) The proposed TB-LSTM model achieved superior performance (R2: 0.853, RMSE: 516.619 kg/ha) compared to the baseline LSTM (R2: 0.353–0.732; RMSE: 735.378–1126.062 kg/ha), confirming its efficiency in resolving feature conflict and better exploiting the yield estimation potential of remote sensing and meteorological data. (2) The integration of meteorological data, spectral reflectance, and vegetation indices proved essential for achieving optimal yield estimation accuracy. Meteorological data provided the most significant contribution, while spectral reflectance and vegetation indices offered complementary information that improved model robustness. When all three data types were utilized simultaneously, the TB-LSTM model achieved peak estimation accuracy (R2: 0.853; RMSE: 514.013 kg/ha; MAE: 380.563 kg/ha). (3) The TB-LSTM model demonstrated robust early prediction capability. Using data from the first 27 time phases (covering growth stages up to heading), it successfully predicted winter wheat yields 48 days before harvest with optimal precision (R2: 0.868; RMSE: 487.327 kg/ha; MAE: 361.353 kg/ha). This capability supports proactive decision-making and resource allocation in agricultural management.