Molecular representation plays a central role in computational drug discovery. Pharmacophores, functional groups responsible for molecular bioactivity, have been widely studied in cheminformatics. However, their incorporation into molecular representation learning, particularly in a context reasoning or generalization, remains relatively limited. To address this gap, we propose PharmaQA, a pharmacophore oriented question answering framework that formulates tailored prompts to extract context-aware molecular semantics. Rather than encoding pharmacophore features, PharmaQA learns to answer pharmacophore related queries. This design enables flexible reasoning across diverse tasks, including molecular property prediction, compound-target interaction prediction, and binding affinity estimation. Experimental results on benchmark datasets demonstrate that PharmaQA achieves competitive performance. In a ligand discovery case study using FDA-approved compounds, the framework identified potential inhibitors for three therapeutic targets, with strong docking performance. As a generalizable and modular solution, PharmaQA incorporates pharmacophoric knowledge into molecular embeddings, enhancing both predictive accuracy and interpretability in drug discovery applications.

In natural scenarios, vision models often encounter the challenge of complex degradation scenarios(e.g., rain, snow, fog, or motion blur). These degradations severely corrupt image features, causing existing models to treat rarely seen or unseen degraded images as “unfamiliar”, thereby losing their inherent recognition and perception capabilities. To address this challenge, we propose a novel degradation disentanglement model (DDM) aimed at precisely disentangling degraded features from the image. The model enhances its perception of various degradations by controlling the matching of features across different degradation types and further strengthens the cross-correlation of target features by introducing a degradation suppression module. This enables the model to re-identify and re-localize targets while removing degradations. We validated the effectiveness of our method on more challenging few-shot segmentation datasets Degraded-Pascal and Degraded-COCO. Results on them outperform SOTA with 3.71% and 3.69% improvement respectively. The experimental results show that our method significantly improves the performance of vision models in various degradation scenarios and provides new ideas and solutions for visual understanding tasks in complex environments.

Drug combinations are widely used in modern medicine but may cause severe adverse drug reactions. Therefore, making effective drug-drug interactions (DDI) prediction is crucial for pharmacovigilance. Existing DDI prediction models are typically built from a structural perspective, assuming that drugs with similar molecular structures may exhibit similar interactions. However, such approaches overlook the biological mechanisms underlying DDI in the human body. This not only weakens the generalization ability of the model, but also makes its interpretability less convincing. Inspired by this, we propose a new method called PC-DDI. Unlike structure-based models, PC-DDI utilizes pharmacophores as basic unit, and designs a complete pharmacophore feature processing framework. It further constructs a pharmacophore-based bipartite graph to model interactions between pharmacophores. This approach allows us to explore the underlying mechanisms of DDI from a functional perspective. We also design a spatial attention weight graph convolution module to optimize the message passing process by integrating pharmacophore position features with node features. Furthermore, we apply causal inference to identify key pharmacophores in pharmacophore bipartite graph, enhancing the interpretability. Compared with the SOTA, PC-DDI achieves an accuracy improvement of 1.84% under the transductive setting and consistently outperforms others in all other experiments.

Vision foundation models (e.g., SAM2, CLIP) show strong generalization in natural image analysis but degrade significantly in specialized domains like medical imaging. This is critical for tasks such as brain tumor segmentation, where errors directly affect surgical planning and patient outcomes. In such contexts, segmentation must be highly reliable and structurally precise, underscoring the need for adaptable methods with low error tolerance. While fine-tuning is the dominant strategy, it is computationally expensive and prone to forgetting. To address this, we propose CausalBridgeNet, a causality-guided correction framework for medical image segmentation. Inspired by predictive coding theories of the Bayesian brain, our method introduces a Predictive Causal Reasoning Unit (PCRU) that estimates structured error maps and delivers targeted feedback to iteratively refine predictions. This forms a closed-loop, error-aware correction mechanism without modifying the foundation model. By keeping the backbone frozen, CausalBridgeNet preserves general visual priors while enhancing task-specific accuracy. On the BraTS 2025 benchmark, it achieves an average Dice score of 84.48 and HD95 of 5.48 across tumor subregions, demonstrating its effectiveness for high-precision medical segmentation.

Effective segmentation of unseen categories in zero-shot semantic segmentation is hindered by models’ limited ability to interpret edges in unfamiliar contexts. In this paper, we propose EdgeCLIP, which addresses this by integrating CLIP with explicit edge-awareness. Based on the premise that edge variation patterns are similar across both seen and unseen class objects, EdgeCLIP introduces the Contextual Edge Sensing module. This module accurately discerns and utilizes edge information, which is crucial in complex border areas where conventional models struggle. Further, our Text-Guided Dense Feature Matching strategy precisely aligns text encodings with corresponding visual edge features, effectively distinguishing them from background edges. This strategy not only optimizes the training of CLIP’s image and text encoders but also leverages the intrinsic completeness of objects, enhancing the model’s ability to generalize and accurately segment objects in unseen classes. EdgeCLIP significantly outperforms the current state-of-the-art method, achieving a deep impressive margin of 17.5% on COCO-20i datasets. Our code is available at github.com/aqingaqinghh/EdgeCLIP.

Molecular representation learning, which captures the fundamental characteristics of chemical compounds, is crucial for AI-driven drug discovery. Existing methods integrate various modalities (e.g., 2D topology and 3D geometry) to develop robust representations. However, current multi-modal fusion strategies either align embedding space through independent models separately, thereby overlooking complementary information, or bridge modalities at a coarse-grained level, failing to capture inherent correlations. We present MolInterAct, an innovative pretraining framework designed to promote multiscale interactions between 2D and 3D modalities at both atomic-level and moleculelevel. Specifically, we propose a fine-grained fusion module, coupled with a customized complementary masking strategy, to seamlessly integrate information at the atomic-level, mitigating overlap and similarity between 2D and 3D representations. In addition, we introduce a fusion contrastive module, which operates at the molecule level, to further strengthen the fusion of 2D and 3D representations while preserving modality-specific features. Finally, we incorporate an intra-modal reconstruction module to reconstruct the original information, further refining the model's understanding of individual modality. Extensive experiments demonstrate that our model outperforms existing molecular pretraining methods across both 2D and 3D benchmarks, highlighting the effectiveness of multiscale fusion between modalities.

Therapeutic peptides demonstrate significant potential in anti-infection, antitumor, and immunomodulation therapies owing to their high specificity and low toxicity. However, existing computational methods are predominantly limited to single-activity design, restricting their clinical applicability. Here, we present MultiPepDec, a novel decoupled prompt learning framework based on protein language model for concurrent generation of multifunctional peptides, which includes antimicrobial, anticancer, toxic, and metabolic activities. Our approach employs: i) Shared-prompts capturing universal therapeutic patterns via adversarial purification; ii) Private-prompts encoding activity-specific knowledge through contrastive learning, ensuring functional decoupling between four activities. Experimental results demonstrate that generated antimicrobial peptides achieve 80.38% predicted efficacy against E. coli, with comparable performance against most clinically relevant pathogens. This confirms robust broad-spectrum capabilities without requiring pathogen-specific training, while maintaining low computational costs. For other therapeutic activities, the designed sequences not only exhibit the intended biological functions but also show significantly improved diversity. This work establishes a new paradigm for efficient multi-activity peptide design, with potential extensions to other biomolecular engineering domains.

Three-dimensional medical image segmentation plays a significant role in clinical diagnosis, treatment planning, and disease research, as it provides doctors with precise anatomical and lesion information and improves the accuracy and efficiency of medical decision-making. However, most existing 3D segmentation approaches rely heavily on densely volumetric data and often fail to perform segmentation properly for incomplete 3D volume acquisition, i.e., missing slices. In this work, we present InterFrameNet, a framework designed to predict intermediate lesion structures by modeling spatial relationships across frames, enabling robust segmentation performance under sparse acquisition conditions, without requiring full-volume information. Our method explicitly models cross-frame spatial continuity and leverages structural relationships between available frames to accurately infer missing lesion regions. This design significantly reduces the dependence on consecutive frames while fully exploiting contextual anatomical information. Extensive experiments on brain lesion datasets demonstrate that our approach achieves robust segmentation performance under sparse acquisition settings, offering a practical solution to maximize usability of incomplete clinical imaging data.

In recent years, significant progress has been made in prototype-based learning methods for few-shot semantic segmentation. However, prototype features originating from the support images are interfered with by intra-class diversity and thus cannot be aligned with the query foreground, resulting in poor segmentation accuracy. Therefore, we propose a novel self-support prototype-aware (SSPA) network to obtain highly confident query foreground pixel points and their corresponding query features. We design Cycle Consistency Collection module and Self-Support Collection module to address the interference of invalid support prototypes. Experimental results demonstrate that our SSPA significantly improves the quality of prototypes and achieves state-of-the-art segmentation results on multiple datasets. In particular, SSPA achieves mIoU scores of 69.7% and 76.4% for 1-shot and 5-shot segmentation, respectively, on PASCAL-5i.

Accurate segmentation of lesions is crucial for disease diagnosis and treatment planning. However, blurring and low contrast in the imaging process can affect segmentation results. We have observed that noninvasive medical imaging shares considerable similarities with natural images under low light conditions and that nocturnal animals possess extremely strong night vision capabilities. Inspired by the dark vision of these nocturnal animals, we proposed a novel plug-and-play dark vision network (DVNet) to enhance the model's perception for low-contrast medical images. Specifically, by employing the wavelet transform, we decompose medical images into subbands of varying frequencies, mimicking the sensitivity of photoreceptor cells to different light intensities. To simulate the antagonistic receptive fields of horizontal cells and bipolar cells, we design a Mamba-Enhanced Fusion Module to achieve global information correlation and enhance contrast between lesions and surrounding healthy tissues. Extensive experiments demonstrate that the DVNet achieves SOTA performance in various medical image segmentation tasks.

Accurately modeling the target-drug complex at atom level presents a significant challenge in the computer-aided drug design. Traditional methods that rely solely on rigid transformations often fail to capture the adaptive interactions between targets and drugs, particularly during substantial conformational changes in targets upon ligand binding, which becomes especially critical when learning target-drug interactions in drug design. Accurately modeling these changes is crucial for understanding target-drug interactions and improving drug efficacy. To address these challenges, we introduce DynaPhArM, an SE(3)-Equivariant Transformer model specifically designed to capture adaptive alterations occurring within target-drug interactions. DynaPhArM utilizes the cooperative scalar-vector representation, drug-specific embeddings, and a diffusion process to effectively model the evolving dynamics of interactions between targets and drugs. Furthermore, we integrate physical information and energetic principles that maintain essential geometric constraints, such as bond lengths, bond angles, van der Waals forces (vdW), within a multi-task learning (MTL) framework to enhance accuracy. Experimental results demonstrate that DynaPhArM achieves state-of-the-art performance with an overall root mean square deviation (RMSD) of 2.01 Å and a sc-RMSD of 0.29 Å while exhibiting higher success rates compared to existing methodologies. Additionally, DynaPhArM shows promise in enhancing drug specificity, thereby simulating how targets adapt to various drugs through precise modeling of atomic-level interactions and conformational flexibility.

Brain tumor segmentation is pivotal in the diagnosis and treatment of brain tumors. As functional imaging technologies like CT and MR advance, analyzing 3D medical image data becomes more time-consuming. Several challenges exist in 3D medical image segmentation: 1) 2D networks, when applied to 3D segmentation tasks, suffer from a lack of 3D structural information. 2) Pure 3D networks, due to their vast parameter count and smaller training sample, are susceptible to overfitting. 3) Current 2.5D networks do not fully leverage the available 3D structural information. In this study, we introduce the Mix-UNet, a multi-branch network that synergizes 2D and 3D networks. This design preserves essential 3D structural details for precise segmentation while ensuring computational efficiency. Our model comprises two main branches and a fusion module: a 2D branch for coarse segmentation without 3D structural information, a 3D branch to capture comprehensive 3D structural details, and a fusion module for pixel-level integration to produce the final segmentation. Experimental results demonstrate the model’s ability to reduce parameter count, increase robustness, and maintain high precision. When tested on the BraTS 2020 validation dataset, our model achieved mean dice coefficients of 90.4%, 80.7%, and 71.2% for the whole tumor, tumor core, and enhancing tumor, respectively, with only 2.2M parameters.

U-shaped network structures with skip connections played an irreplaceable role in medical image analysis, but the limitation of convolution makes it unable to learn long-distance semantic information well. The recent success of Transformer in natural language processing and image classification shows that it can benefit from global information modeling by using self-attention mechanisms. However, both local and global features are equally important for dense prediction tasks. Transformer ignores local semantic information to a certain extent. In this study, we propose a Unet-like Transformer for medical image segmentation, named MN-Unet, which can simultaneously extract local and global features. MN-Unet consists of encoder, decoder, and skip connections. Specially, we design an encoder based on the Neighborhood Attention Transformer, which fuse three neighborhood sizes of different dimensions to simultaneously extract local and global features. In the decoder, we use bilinear interpolation to restore the image to its original size. Skip connection is added to alleviate the distortion of low resolution to high resolution. MN-Unet can achieve accurate segmentation of medical images without any pre-training. Extensive experimental results on two medical image datasets (LiTS 2017 and BraTS 2020) show that we achieve relatively better performance than state-of-the-art methods. The codes and trained models will be publicly available a https://github.com/hutchinsonian/MN_Unet

MOTIVATION: Recently, with the development of high-throughput experimental technology, reconstruction of gene regulatory network (GRN) has ushered in new opportunities and challenges. Some previous methods mainly extract gene expression information based on RNA-seq data, but the associated information is very limited. With the establishment of gene expression image database, it is possible to infer GRN from image data with rich spatial information. RESULTS: First, we propose a new convolutional neural network (called SDINet), which can extract gene expression information from images and identify the interaction between genes. SDINet can obtain the detailed information and high-level semantic information from the images well. And it can achieve satisfying performance on image data (Acc: 0.7196, F1: 0.7374). Second, we apply the idea of our SDINet to build an RNA-model, which also achieves good results on RNA-seq data (Acc: 0.8962, F1: 0.8950). Finally, we combine image data and RNA-seq data, and design a new fusion network to explore the potential relationship between them. Experiments show that our proposed network fusing two modalities can obtain satisfying performance (Acc: 0.9116, F1: 0.9118) than any single data. AVAILABILITY AND IMPLEMENTATION: Data and code are available from https://github.com/guofei-tju/Combine-Gene-Expression-images-and-RNA-seq-data-For-infering-GRN.

Fast and accurate segmentation of medical images is essential for the diagnosis and treatment of diseases. The automatic segmentation technology based on deep learning has achieved encouraging performance in segmentation accuracy. However, the improvement of segmentation accuracy usually requires a larger network structure, which also leads to a decrease in segmentation speed. In this study, we propose a medical image anticipation segmentation net (MIASNet), in order to further improve the segmentation speed under the premise of excellent segmentation accuracy. For 3D medical images, we use the spatial association of the previous frame and the current frame as input data to predict the segmentation results of the next frame. Our approach consists of three lightweight sub-networks, which are used to learn the mapping relationship of the spatial domain. In order to make full use of the generating ability of the deep learning network, we use group convolution to obtain diversified prediction results. On the multimodal brain tumor image segmentation (BraTS) 2020 dataset, MIASNet achieves excellent segmentation accuracy without using the target frame that need to be segmented as the network input. Therefore, our proposed segmentation network can be used in a wider range of real-time medical applications.

With the wide application of deep learning technology in medical image processing, the performance of medical image segmentation has been improved in a breakthrough. U-Net architecture has excellent performance in medical image segmentation tasks. In order to solve the problem of image signal loss caused by the autoencoder structure, U-Net has added skip connections to its network to transfer the low-level features of the encoder path to the decoder path. Although this method can roughly solve the problem of image information loss, while it introduces a new problem, that is, the simple feature fusion method causes the high-level semantic information to be diluted. In order to solve the problem that the simple fusion of low-level edge information and high-level semantic information creates the semantic gap and dilutes high-level semantic information, we propose a novel U-shaped architecture, namely GEU-Net. GEU-Net utilizes ensemble learning methods to obtain better segmentation performance with a small computational cost. In addition, We propose a multi-scale group convolution block namely Group Residual (GR) module to reduce the semantic gap between encoder and decoder. We have evaluated our model on the BraTS 2020 Challenge, and have achieved competitive segmentation results.

Accurate automatic medical image segmentation technology plays an important role for the diagnosis and treatment of brain tumor. However, existing methods based on outstanding 2.5D and 3D segmentation strategies are time-consumption and hardware-consumption while ensuring high accuracy. In order to reduce the high demand for automatic segmentation of tumor images and avoid the noise interference in a single input image, we propose an end-to-end zero-shot CNN segmentation method. Our method only utilizes two adjacent images, instead of the target image, as the input data of deep neural network to predict the brain tumor area in the target image. Avoiding noise interference in the target image, this method makes full use of the spatial context feature between adjacent slices in order to obtain accurate zero-shot segmentation results. We compare with the state-of-the-art segmentation frameworks on the same benchmark and notice that our method has strong competitiveness.

Auotmatic melanoma classification in dermoscopy images is a very important task, which can help improve diagnostic accuracy and reduce mortality. Deep convolutional neural network (DCNN) has developed rapidly in recent years, but it is still a challenging task due to the intra-class variation and inter-class similarity of melanoma. We proposed a novel neural network integration model, which is composed of three parts: First, we use U-net segmentation network to generate masks and use the masks to crop original images; Second, we use five state-of-the-art DCNNs to extract features of cropped images, and add the squeeze-excitation block (SE block) to emphasize useful features; Finally, we construct a new neural network with local connection to integrate the classification results, extract features of different class of results, and integrate the results of each class separately. Local connection can integrate each class separately, maximizing the advantages of different networks in various classes. We evaluate our model on ISIC 2017 challenge dataset, and the result shows that our method has better performance compared with the existing methods.

The sea has been manifested to cause the emotion of fear to people when it comes to a very depth, especially to those who have thalassophobia. Many people have to work in the sea while nearly no research on influence of fear of the sea to cognition has been carried out. This study explores the impact of fear of the sea induced by immersive virtual reality on working memory which is a cognitive system with a limited capacity. Participants were required to complete n-back working memory task of three difficulty levels in the non-emotional environment and the undersea environment respectively by means of virtual reality. Pupil diameter changes were recorded along with the task performance. In addition to reaction times and accuracy (correctly press a button in response to targets) as two task performance indices used in most researches, the commission errors (incorrectly press a button in response to non-targets) and omission errors (incorrectly do not press a button in response to targets) were also differentiated herein. The results of the study indicated that the virtual undersea environment did induce the emotion of fear. As for the task performance, except that the performance of low-level task did not differ much between the two environments, the fear of the sea increased the accuracy of the medium level n-back task but decreased it of high-level n-back task. Result of omission errors was just the opposite and commission errors were increased in both levels of task. The findings, including the positive role of a moderate level of fear of the sea in the performance of working memory task, make a lot of sense for future cognitive work in the sea.