Interpreting the functional consequences of coding variants remains a central challenge in human genetics, particularly given the clinical importance of distinguishing pathogenic mutations from benign variation. Here we present IPVAR, a deep learning framework that uniquely integrates tertiary protein structural features predicted by both ESM C and AlphaFold 3, alongside established conservation metrics, to advance variant pathogenicity prediction. IPVAR leverages a hierarchical crossattention mechanism to capture both global and fine-grained structural dependencies between complementary representations, and incorporates an adaptive modality weighting strategy to dynamically balance information from each protein structure model. Comprehensive benchmarking demonstrates that IPVAR substantially outperforms state-of-the-art methods, including those based solely on sequence annotations or individual structural predictors, achieving an area under the ROC curve (AUC) of 0.9838 on the ClinVar dataset and 0.9202 on an independent Mendelian disease variant cohort. Ablation studies further confirm that both the multi-model integration and advanced fusion modules are critical to the model's superior performance. These findings establish IPVAR as a new benchmark for the functional interpretation of genomic variants, and highlight the value of integrating diverse structural foundation models to improve clinical variant assessment.

Cancer is heterogeneous, affecting the precise approach to personalized treatment. Accurate subtyping can lead to better survival rates for cancer patients. High-throughput technologies provide multiple omics data for cancer subtyping. This study proposed Subtype-Former, a deep learning method based on MLP and Transformer Block, to extract the lowdimensional representation of the multi-omics data. K-means and Consensus Clustering are also used to achieve accurate subtyping results. We compared Subtype-Former with the other state-of-the-art subtyping methods across the TCGA 10 cancer types. We found that Subtype-Former can perform better on the benchmark datasets of more than 5000 tumors based on the survival analysis. In addition, Subtype-Former also achieved outstanding results in pan-cancer subtyping, which can help analyze the commonalities and differences across various cancer types at the molecular level. Finally, we applied Subtype-Former to the TCGA 10 types of cancers. We identified 50 essential biomarkers, which can be used to study targeted cancer drugs and promote the development of cancer treatments in the era of precision medicine.

In early lung cancer screening, precise classification of benign and malignant pulmonary nodules is of critical importance to clinical decision making and individualized treatment. Although early methods have achieved considerable progress, two main problems remain: existing diagnostic methods have limitations in utilizing multimodal data and capturing semantic information, and traditional multimodal approaches relying on late-stage feature fusion fail to facilitate the valuable information from internal model layers. To overcome these challenges, we propose Multimodal and Multilevel Merging Net (MMMNet), a multimodal architecture for pulmonary nodule classification that includes two innovative modules: a Multimodal Feature Fusion Module that combines computed tomography scans and text annotations parallelly in multiple layers to construct a feature pyramid, and a Multilevel Feature Merge Module that recursively merges the fused features to utilize both high-level semantic information and low-level visual characteristics. The approach also integrates Focal Loss to tackle the imbalanced classification and Contrastive Loss to align the multimodal features. The proposed approach is evaluated on the LIDC-IDRI dataset, yielding an accuracy of$\mathbf{9 1. 7 3 \%}$and a specificity of$\mathbf{9 5. 5 2 \%}$. Experiment results show that the approach enhances the indepth multimodal feature mining and has a promising potential in medical image analysis and clinical application.

Gas recognition under few-shot conditions presents several challenges, primarily due to sensor selectivity and the complexities of data collection. While meta-learning techniques have shown promise in improving adaptability to various gas types, sensor configurations, and environmental conditions with limited data, limitations remain. To address these issues, researchers employ diverse sensor arrays, supported by neural networks that capture complex data dependencies. However, gas recognition in few-shot scenarios is still in its early stages, highlighting the need for further research. In this paper, we propose an innovative meta-learning-based framework based on task enhancement and global-local information fusion for few-shot gas classification. Firstly, to enhance data utilization and generalization capabilities, we propose the Global and Local Information Fusion (GLIF). Specifically, GLIF has two modules: the Global Information Extraction Module (GIEM) and the Local Information Extraction Module (LIEM). GIEM captures long-range dependencies in the gas signal, while LIEM focuses on detailed local features using convolutional layers, reducing model complexity and the risk of overfitting. Secondly, to improve performance in challenging tasks, we design a Special Task Enhancement Module (STEM). Concretely, STEM identifies the most challenging categories, aggregates tasks containing these categories, and resamples them for focused retraining based on a ranking-based strategy. Experimental results validate the effectiveness of our proposed method, demonstrating high recognition accuracy and robust generalization ability. These traits are particularly evident in scenarios involving gas drift and limited samples, highlighting the potential value and impact of our approach in real-world applications.

Driver genes play a crucial role in the growth of cancer cells. Accurate identification of cancer driver genes is essential for deepening our understanding of cancer pathogenesis and facilitating the development of cancer therapies and drug-targeted driver genes. However, the diversity and complexity of multi-omics data still make cancer driver identification highly challenging. In this study, we propose Transformer-Driver (Trans-Driver), a deep supervised learning method based on a novel transformer architecture, which integrates multi-omics data to learn the differences and associations between different omics modalities for cancer driver discovery. Trans-Driver introduces a kernel-based multi-head self-attention mechanism with gated residual connections, as well as a Dynamic Tanh (DyT) normalization function, to enhance the integration and modeling of heterogeneous multi-omics features. Compared with other state-of-the-art driver gene identification methods, Trans-Driver achieved excellent performance on TCGA, CGC, and PCAWG datasets. Among approximately 20,000 protein-coding genes, Trans-Driver reported 269 candidate driver genes, of which 132 genes (about 49.1%) were included in the gold standard CGC dataset. Feature contribution analysis further demonstrated that integrating multi-omics data improved performance compared to using somatic mutation data alone. Finally, detailed analysis revealed that the candidate drivers are clinically meaningful, demonstrating the practical value of Trans-Driver.

Few-shot class-incremental learning (FSCIL) requires learning new knowledge without forgetting old knowledge. Forward compatibility can reserve space for novel classes while maintaining base class knowledge in incremental learning. Better forward compatibility is crucial for effectively mastering all knowledge, especially when dealing with a few unknown new classes. In this article, we propose the better forward compatibility pretraining (BFCP) to further enhance forward compatibility in FSCIL. We adopt a two-stage training for the backbone network in the base session. First, we train the backbone network at the image-level to enhance its feature extraction capability, enabling the model to extract valuable information from unknown class images. Second, we fine-tune the backbone network at the feature-level with fake prototypes and instances to achieve clustering base classes and reserve space for unknown new classes. For all incremental new sessions, we freeze the backbone network and employ prototype rectification without further training to refine the prototypes of the novel classes. We conduct extensive experiments with different input scales, including federated cross-domain pretraining and cross-domain class-incremental experiments. BFCP efficiently handles both novel and base classes of each incremental session and significantly outperforms state-of-the-art methods, achieving an average accuracy of 63.47% on the CIFAR100 dataset.

Accurate detection of depression through speech analysis offers a promising non-invasive approach for early diagnosis and intervention. However, the high dimensionality and complexity of speech features present significant challenges in identifying the most relevant features for depression detection. This study applies the spFSR (Feature Selection and Ranking via Simultaneous Perturbation Stochastic Approximation) technique to a comprehensive 2,268-dimensional speech feature set, focusing on selecting features specifically relevant to depression. The effectiveness of the spFSR method is evaluated using two well-known datasets: DAIC-WOZ and CMDC. The selected feature set was assessed across various machine learning models, demonstrating substantial improvements in key performance metrics on both datasets. The results indicate that the spFSR method effectively optimizes feature selection for depression detection, leading to more robust and accurate predictive models across different datasets. Our study find that the top 10 features for effective speech-based depression detection include spectral features (e.g., spectral flux, entropy, flatness), fundamental frequency metrics (e.g., lowest percentile), periodic features (e.g., jitter), and MFCC attributes (e.g., segment length, skewness).

With the ongoing advancements in science and technology and the increasing research focus on cancer-related issues, there has been a proliferation of omics-related resources for in-depth analysis and exploration. This burgeoning volume and complexity of biological data have fostered the integration of machine-learning techniques into biology. As a result, numerous machine-learning strategies have been established to identify driver mutations. Yet, many of these strategies produce complex models, complicating comprehension and thereby clouding the impact of input features on the resulting predictions. Our analysis presented the CIXG framework, which integrates a driver gene prediction module using XGBoost with a causality interpretation module anchored on CXPlain. This architecture enables quantifying each input feature’s contribution to the prediction outcome and ensures precise predictions of driver genes. When benchmarked against the state-of-the-art (SOTA) method, CIXG demonstrated superior accuracy in pinpointing driver genes across pan-cancer studies and within the 32 specific cancer types. Importantly, our results underscored that mutation features chiefly influence CIXG’s predictive prowess, with additional support from other omics features.

Infrared dim small target (IDST) detection is of great significance in security surveillance and disaster relief. However, the complex background interference and tiny targets in infrared images keep it still a long-term challenge. Existing deep learning models stack network layers to expand the model fitting capability, but this operation also increases redundant features which reduce model speed and accuracy. Meanwhile, small targets are more susceptible to positional bias, with this dramatically reducing the model’s localization accuracy. In this article, we propose a guided attention and joint loss (GA-JL) network for infrared small target detection. More specifically, the method visualizes the feature maps at each resolution through a two-branch detection head (TDH) module, filters out the features that are strongly related to the task, and cuts out the redundant features. On this basis, the guided attention (GA) module guides the prediction layer features using the features that are associated closely with the task and combines spatial and channel bidirectional attention to make the prediction layer feature maps embedded with effective messages. Finally, through the joint loss (JL) module, the target position regression is performed with multiangle metrics for enhancing the target detection accuracy. Experimental results of our method on the SIATD, SIRST, and IRSTD_1k datasets reveal that it is capable of accurately identifying IDSTs, remarkably reduces the false alarm rate, and outperforms other methods.

In the context of long-range infrared detection of small targets, complex battlefield environments with strong background effects, adverse weather conditions, and intense light interference pose significant challenges. These factors contribute to a low signal-to-noise ratio and limited target information in infrared imagery. To address these challenges, a feature enhancement network called target-focused enhancement network (TENet) is proposed with two key innovations: the dense long-distance constraint (DLDC) module and the autoaugmented copy-paste bounding-box (ACB) strategy. The DLDC module incorporates a self-attention mechanism to provide the model with a global understanding of the relationship between small targets and the backgrounds. By integrating a multiscale structure and using dense connections, this module effectively transfers global information to the deep layers, thus enhancing the semantic features. On the other hand, the ACB strategy focuses on data enhancement, particularly increasing the target representation. This approach addresses the challenge of distributional bias between small targets and background using context information for target segmentation, mapping augment strategies to 2-D space, and using an adaptive paste method to fuse the target with the background. The DLDC module and the ACB strategy complement each other in terms of features and data, leading to a significant improvement in model performance. Experiments on infrared datasets with complex backgrounds demonstrate that the proposed network achieves superior performance in detecting dim and small targets.

Infrared Dim Small Target (IDST) detection holds significant importance in early target warning and ground monitoring. However, IDST detection remains a long-standing challenge due to the low signal-to-noise ratio and low contrast. Feature fusion is an effective approach for feature enrichment and improving poor performance in IDST detection. Existing feature fusion methods tend to overlook the importance of focusing on both multi-layer and single-layer features, which prevents target features from being fully exploited, resulting in suboptimal outcomes for IDST detection. In this paper, we present a Multi-Angle Pyramid Feature Fusion Network (MAPFF), which selects fusion objects from multiple perspectives and then fuses them. Namely, multilayer features and single-layer features - two perspectives of the fusion object - are selected and fused separately. The MAPFF network consists of two primary modules: a Cross-Layer Complementary Feature (CLCF) module and an Atrous Spatial Pyramid Pooling with Attention (AttnASPP) module. To effectively fuse semantic and geometric detail information, the CLCF module adaptively combines different layer features as complementary features, while the original layer features serve as the main features. Concurrently, through channel shuffle, the complementary and main features achieve substantial information exchange. The AttnASPP module employs parallel atrous convolutions with multiple dilation rates to obtain multi-scale information and incorporates an attention mechanism to emphasize effective features. Experimental results on the SIATD, SIRST and IRSTD_1k datasets demonstrate that our method can precisely identify IDSTs, significantly reduce the false alarm rate, and outperform other methods.

Inspired by the well-known Papez circuit theory and neuroscience knowledge of reinforcement learning, a double dueling deep Q network (DQN) is built incorporating the electroencephalogram (EEG) signals of the frontal lobe as prior information, which is named frontal lobe double dueling DQN (FLD3QN). The framework of FLD3QN is constructed in accord with the brain emotion mechanism which takes the frontal lobe and the thalamus as the core, in which the part of the Papez circuit is simulated by the bifrontal lobe residual convolution neural network (BiFRCNN). Moreover, a step penalty factor is designed to constrain the number of mistakes of the agent. The ablation studies results on the public EEG emotion dataset DEAP verified the important roles of the frontal lobe and the Papez circuit in modeling the procedure of learning rewards during the perception of emotions, with a great increase in the average accuracies by 25.24% and 23.31% in valence and arousal dimensions.

Advancements in single-cell multi-omics sequencing technologies have dramatically transformed the analysis of cellular states at single-cell resolution. Cluster analyses leveraging single-cell transcriptome and epigenome data have enabled the characterization of cellular states and the delineation of transcriptomic regulatory programs associated with cellular heterogeneity. However, the high dimensionality, sparsity, and heterogeneity inherent to multi-omics data present significant challenges to their clustering analysis. In this study, we introduced single-cell Multi-Omics Generative Adversarial Networks (scMOGAN), an unsupervised clustering model specifically tailored for single-cell transcriptome and epigenome data. scMOGAN accomplishes this by efficaciously modeling the omics data and mining potential features through neural networks, ultimately utilizing Gaussian mixture models for cell type identification. To validate and scrutinize the performance of scMOGAN, we performed analyses on two actual multi-omics datasets and one simulated dataset. The results demonstrate that scMOGAN surpasses other existing methods in terms of clustering performance, underscoring its efficacy in navigating the complexities of multi-omics data.

Genomic variants, which can disrupt cellular functions, present a challenge in distinguishing deleterious from benign instances. While assessing genome-wide functional impacts, many current algorithms neglect protein tertiary structure of coding region variants due to limitations in protein structural prediction. This study introduces PMMVar, an advanced multimodal deep convolutional network, which adeptly integrates protein tertiary structures with conservation properties from ESM-2, supplemented by other protein structural sequences. PMMVar achieves outstanding performance on the latest clinical variant datasets, NCBI ClinVar (2023), and the Mendelian variant dataset, surpassing existing benchmarks. Ablation analyses validate the significance of protein multi-level structures in enhancing the model’s accuracy. Overall, our findings spotlight the essential role of multi-level protein structures in pathogenicity predictions and their potential to discern deleterious genomic variants effectively.

Hepatocellular Carcinoma (HCC) holds a record of high incidence and severe global harm. In tasks of liver cancer segmentation based on 3D medical images, the majority of methods have endeavored to enhance the 3D U-net by integrating the latest modules from the field of Computer Vision (such as the transformer), often overlooking the distinct characteristics of liver components. We introduced a novel deep learning architecture, Modif-SegUnet, to circumvent this limitation. This architecture extends the U-net model by incorporating a 3D-modifiable attention module, thereby fostering a heightened focus on distinguishing between normal liver sections and lesions. Furthermore, Modif-SegUnet ingeniously amalgamates the 3D-modifiable attention module with the Modif-transformer block, enabling efficient capture of relevant and valuable full-text information in CT images that contain liver or tumor regions. We subjected the proposed Modif-SegUnet to evaluation on the Liver Tumor Segmentation benchmark dataset. Experimental outcomes indicate that our methodology surpasses state-of-the-art approaches in liver tumor segmentation, suggesting potential pathways for advancing the diagnosis and treatment of liver cancer.

The COVID-19 pandemic has had a profound impact on human society. It has highlighted the need for faster diagnostic methods. Research has shown that combining semantic segmentation with traditional medical approaches can significantly accelerate the process. To address this, leveraging COVID-19 CT images, our team designs a revolutionary semantic segmentation model called Level of Detail Enhancement U-Net (LDE-UNet), which shows the lesion area on CT images. By introducing the LDE block, the model has the unique advantage of overcoming the loss of data details during the downsampling process by emphasizing and transmitting details at the same level. Our SOTA model outperforms the second-best model by at least 0.7% in the most critical indicator precision. Compared with other models, LDE-UNet’s strong reliability determines its ability to be used in the medical field to accelerate the localization and division of lesion areas on CT images by professional doctors, thus completing patient diagnosis faster. In addition, we also propose a standardized method for processing medical images.

Cancer survival prediction is pivotal in tailoring individualized treatment strategies and guiding clinician decision-making. Yet, existing methodologies grapple with efficiently harnessing the intricate distribution of medical data spanning various modalities. In response, we present SAMMS, an advanced multi-omics multimodal deep learning framework tailored for survival prediction. SAMMS leverages the robust image segmentation model, "Segment Anything" to adeptly characterize pathological images. This prowess is further enhanced by integrating multi-omics data and clinical insights, facilitating holistic modeling across a diverse modal spectrum. The framework weaves a modality-specific subnetwork with a cross-modality common subnetwork, meticulously capturing intra-modality nuances and inter-modality correlations. SAMMS eclipsed its contemporaries by delivering remarkable performance on TCGA’s LGG and KIRC tumor datasets. A battery of analyses underscored SAMMS’s unparalleled capability to distill multifaceted insights from multimodal datasets, yielding richer and more integrative multimodal representations. Such strides promise significant advancements in cancer survival analytics, bolstering the precision and efficacy of patient-centric treatments, disease oversight, and clinical decision processes.

The strength of emotion at different positions in a speech is strong or weak, and the weak parts with unclear emotions will bring noise to the model. We propose a boosting ensemble learning method based on mixed entropy down-sampling to effectively select emotionally salient segments to improve the classifier's performance. An independent Convolutional Neural Network (CNN) model is trained in each iteration of ensemble learning. These CNN models form an ensemble classifier, which improves the generalization ability by synthesizing all the learned results of down-sampling, making emotion recognition more accurate. We also introduce the concept of Mixed Information Entropy (MIE), which consists of Emotional Certainty Entropy (ECE) and Structural Distribution Entropy (SDE). ECE measures the emotional confusion of segments, while SDE measures the stability of segments in deep feature space. During the iteration, the deep features are obtained from the last fully connected layer of the model and down-sampled according to the weighted sum of confidence and MIE. The selected segments with stronger emotions are used for the next iteration. Our method is 3.77% higher on WA and 2.37% higher on UA than the naive CNN model on the IEMOCAP dataset.

In the real scene, changes in speaker features caused by different emotional states have a great impact on the performance of speaker recognition. To improve the robustness of the speaker recognition system, the existing emotional speaker recognition technologies tend to cascade different models, ignoring the frame-level acoustic features and the segment-level discourse habits feature. To this end, we combine frame- and segment-level features in different ways to build a robust recognition system for emotional speakers. The frame-level features and segment-level features are jointly learned to retain emotional information and speaker information. Four joint learning methods, namely, Joint in series, Joint in Parallel, Joint under the guidance, and Joint with Original Feature, are discussed to explore the correlations between fragment-level features and frame-level features. The experimental results illustrate that the speaker feature will change greatly in different emotional states. Compared with the accuracy of 90.95% by x-vector, the proposed methods of Joint in parallel and Joint with Original Features can achieve the accuracy of 95.06% and 94.67% respectively for emotional speaker recognition in the experiment on Mandarin Affective Speech Corpus (MASC). Our findings provide a novel aspect to improve speaker recognition robustness.

The development of targeted drugs allows precision medicine in cancer treatment and optimal targeted therapies. Accurate identification of cancer druggable genes helps strengthen the understanding of targeted cancer therapy and promotes precise cancer treatment. However, rare cancer-druggable genes have been found due to the multi-omics data's diversity and complexity. This study proposes deep forest for cancer druggable genes discovery (DF-CAGE), a novel machine learning-based method for cancer-druggable gene discovery. DF-CAGE integrated the somatic mutations, copy number variants, DNA methylation and RNA-Seq data across ˜10 000 TCGA profiles to identify the landscape of the cancer-druggable genes. We found that DF-CAGE discovers the commonalities of currently known cancer-druggable genes from the perspective of multi-omics data and achieved excellent performance on OncoKB, Target and Drugbank data sets. Among the ˜20 000 protein-coding genes, DF-CAGE pinpointed 465 potential cancer-druggable genes. We found that the candidate cancer druggable genes (CDG) are clinically meaningful and divided the CDG into known, reliable and potential gene sets. Finally, we analyzed the omics data's contribution to identifying druggable genes. We found that DF-CAGE reports druggable genes mainly based on the copy number variations (CNVs) data, the gene rearrangements and the mutation rates in the population. These findings may enlighten the future study and development of new drugs.

Cancer driver genes are critical in driving tumor cell growth, and precisely identifying these genes is crucial in advancing our understanding of cancer pathogenesis and developing targeted cancer drugs. Despite the current methods for discovering cancer driver genes that mainly rely on integrating multi-omics data, many existing models are overly complex, and it is difficult to interpret the results accurately. This study aims to address this issue by introducing InDEP, an interpretable machine learning framework based on cascade forests. InDEP is designed with easy-to-interpret features, cascade forests based on decision trees and a KernelSHAP module that enables fine-grained post-hoc interpretation. Integrating multi-omics data, InDEP can identify essential features of classified driver genes at both the gene and cancer-type levels. The framework accurately identifies driver genes, discovers new patterns that make genes as driver genes and refines the cancer driver gene catalog. In comparison with state-of-the-art methods, InDEP proved to be more accurate on the test set and identified reliable candidate driver genes. Mutational features were the primary drivers for InDEP's identifying driver genes, with other omics features also contributing. At the gene level, the framework concluded that substitution-type mutations were the main reason most genes were identified as driver genes. InDEP's ability to identify reliable candidate driver genes opens up new avenues for precision oncology and discovering new biomedical knowledge. This framework can help advance cancer research by providing an interpretable method for identifying cancer driver genes and their contribution to cancer pathogenesis, facilitating the development of targeted cancer drugs.

In the age of Internet of Things where information is explosively growing, people pay more attention on personal privacy. In the real-world task-incremental scenario for biometrics, every edge device faces continuous task flows of private data without communication with others. security and performance are the primary concerns in identity authentication, and federated continual learning (FCL) is a promising solution. In this article, we design a personalized FCL framework to solve the problem of sequential identification in every distributed device. For each client, we create an adaptive continual metalearning model called continual task-distillation-based adaptive model-agnostic metalearning (cTD-$\alpha $MAML), aiming to align the gradients of previous and new tasks and to make the learning-rate (LR) model learnable. For central aggregation, the server gathers the metainitialization from every local update and allocates the updated global metainitialization to clients. We propose an extension of federated average to locally reserve the learnable LR network to realize the personalization of clients. Results prove that in continual learning, our cTD-$\alpha $MAML can learn to learn the seen tasks and avoid catastrophic forgetting. And in FCL, our personalized method realizes the knowledge transferring across clients, meanwhile improving the local performance and reducing the communication cost. In this way, the proposed personalized FCL framework can obtain a biometric template that is able to learn the expression space for new tasks with rapid adaption.

Federated Continual Learning (FCL) approaches exist two major problems of the probability bias and the imbalance in parameter variations. These two problems lead to catastrophic forgetting of the network in the FCL process . Therefore, this paper proposes a novel FCL framework, Federated Probability Memory Recall (FedPMR), to mitigate the probability bias problem and the imbalance in parameter variations. Firstly, for the probability bias problem, this paper designs the Probability Distribution Alignment (PDA) module, which consolidates the memory of old probability experience. Specifically, PDA maintains a replay buffer and uses the probability memory stored in the buffer to correct the offset probabilities of the previous tasks during the two-stage training. Secondly, to alleviate the imbalance in parameter variations, this paper designs the Parameter Consistency Constraint (PCC) module, which constrains the magnitude of neural weight changes for previous tasks. Concretely, PCC applies a set of adaptive weights to subsets of the regularization term that constrains parameter changes, forcing the current model to be sufficiently close to the past model in parameter space distance. Experiments with various levels of task similitude across clients demonstrate that our technique establishes the new state-of-the-art performance when compared to previous FCL approaches.

In infrared scenes, humans can easily observe objects in the scene with their eyes, even dim ones. To make the robot have the same visual ability, this paper proposes a pyramid-feature fusion target detection network, called RLPGB-Net, which combines reinforcement learning with aerial targets in the infrared scene. It makes use of the powerful decision-making ability of reinforcement learning to give corresponding weights to the extracted features and highlight the significant features of infrared dim small targets. In reinforcement learning, we use priori strategy guidance and long-term training methods to train weight-regulating agents. To eliminate the local influence on the detection results, such as bright interference points similar to the target, and to solve the problem of dim target detection effectively, the global context boundary attention module is introduced to eliminate the disadvantage of local comparison by using the global characteristics of different dimensions. At the same time, it can prevent the edge information of the refined target from being submerged in the background. Experimental results on SAITD and SIRST data sets show the effectiveness of the proposed method.

In multilabel images, the changeable size, posture, and position of objects in the image will increase the difficulty of classification. Moreover, a large amount of irrelevant information interferes with the recognition of objects. Therefore, how to remove irrelevant information from the image to improve the performance of label recognition is an important problem. In this article, we propose a convolutional network based on feature denoising and details supplement (FDDS) to address this issue. In FDDS, we first design a cascade convolution module (CCM) to collect spatial details of upper features, in order to enhance the information expression of features. Second, the feature denoising module (FDM) is further put forward to reallocate the weight of the feature semantic area, in order to enrich the effective semantic information of the current feature and perform denoising operations on object-irrelevant information. Experimental results show that the proposed FDDS outperforms the existing state-of-the-art models on several benchmark datasets, especially for complex scenes.

Recently, electroencephalogram (EEG) emotion recognition has gradually attracted a lot of attention. This brief designs a novel frame-level teacher-student framework with data privacy (FLTSDP) for EEG emotion recognition. The framework first proposes a teacher-student network without prior professional information for automated filtering of useful frame-level features by a gated mechanism and extracting high-level features by using knowledge distillation to capture the results of EEG emotion recognition from a teacher network and student networks. Then, the results from subnetworks are integrated by using the novel decision module, which, motivated by the voting mechanism, adjusts the composition of feature vectors and improves the weight of accurate prediction to optimize the integration effect. During training, an innovative data privacy protection mechanism is applied for avoiding data sharing, where each student network only inherits weights from all trained networks and does not inherit the training dataset. Here, the framework can be repeatedly optimized and improved by only training the next student subnetwork on new EEG signals. Experimental results show that our framework improves the accuracy of EEG emotion recognition by more than 5% and gets state-of-the-art performance for EEG emotion recognition in the subject-independent mode.

The grim situation of novel coronavirus pneumonia 2019 (COVID-19) and its terrible spreading speed have already constituted a severe risk to human life, so it is ultimately essential to rapidly and accurately diagnose for COVID-19 pneumonia. Based on this study’s 746 lung CT images, we propose Multi-MedVit, a novel auxiliary COVID-19 diagnosis framework based on the multi-input Transformer. We compare Multi-MedVit with state-of-the-art deep learning methods, such as CNN, VGG16, and ResNet50. Multi-MedVit outperformed the other methods on the benchmark dataset and proved that multiscale data input for data augmentation helped enhance model stability. Based on an interpretable analysis of the input and output of Multi-MedVit, we found that with the support of the training set data, the model has been possible to accurately focus on the lesion area for diagnosis of COVID-19 without expert annotations, which can provide initial references containing more potential information to doctors more precisely and fleetly.

With the development of next-generation sequencing technology, massive genomic data has been generated, primarily encouraging research on cancer driver genes. Many bioinformatics methods were proposed to identify driver genes. However, the results of driver gene identification a mong these methods show considerable differences. It is still challenging to obtain a comprehensive catalog of cancer drivers. Although current methods have greatly promoted the development of driver genes, few methods can integrate the identification results of existing methods. To solve such problems in cancer driver genes research, we proposed a cascade forest model to discover cancer driver genes(CFC) that can integrate multi-omics data and annotation scores from different cancer driver gene identification algorithms. The proposed method got precise results for 33 cancer types and Pan-cancer. The CFC framework identified 275 driver genes in Pan-cancer, of which 179 were included in the Gold standard. The identified genes were enriched i n t he principal cancer signaling pathways.

MOTIVATION: Analysis of whole-genome sequencing (WGS) for genetics is still a challenge due to the lack of accurate functional annotation of non-coding variants, especially the rare ones. As eQTLs have been extensively implicated in the genetics of human diseases, we hypothesize that rare non-coding variants discovered in WGS play a regulatory role in predisposing disease risk. RESULTS: With thousands of tissue- and cell-type-specific epigenomic features, we propose TVAR. This multi-label learning-based deep neural network predicts the functionality of non-coding variants in the genome based on eQTLs across 49 human tissues in the GTEx project. TVAR learns the relationships between high-dimensional epigenomics and eQTLs across tissues, taking the correlation among tissues into account to understand shared and tissue-specific eQTL effects. As a result, TVAR outputs tissue-specific annotations, with an average AUROC of 0.77 across these tissues. We evaluate TVAR's performance on four complex diseases (coronary artery disease, breast cancer, Type 2 diabetes and Schizophrenia), using TVAR's tissue-specific annotations, and observe its superior performance in predicting functional variants for both common and rare variants, compared with five existing state-of-the-art tools. We further evaluate TVAR's G-score, a scoring scheme across all tissues, on ClinVar, fine-mapped GWAS loci, Massive Parallel Reporter Assay (MPRA) validated variants and observe the consistently better performance of TVAR compared with other competing tools. AVAILABILITY AND IMPLEMENTATION: The TVAR source code and its scores on the ClinVar catalog, fine mapped GWAS Loci, high confidence eQTLs from GTEx dataset, and MPRA validated functional variants are available at https://github.com/haiyang1986/TVAR. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Multiple empirical kernel learning (MEKL) is a scalable and efficient supervised algorithm based on labeled samples. However, there is still a huge amount of unlabeled samples in the real-world application, which are not applicable for the supervised algorithm. To fully utilize the spatial distribution information of the unlabeled samples, this paper proposes a novel semi-supervised multiple empirical kernel learning (SSMEKL). SSMEKL enables multiple empirical kernel learning to achieve better classification performance with a small number of labeled samples and a large number of unlabeled samples. First, SSMEKL uses the collaborative information of multiple kernels to provide a pseudo labels to some unlabeled samples in the optimization process of the model, and SSMEKL designs pseudo-empirical loss to transform learning process of the unlabeled samples into supervised learning. Second, SSMEKL designs the similarity regularization for unlabeled samples to make full use of the spatial information of unlabeled samples. It is required that the output of unlabeled samples should be similar to the neighboring labeled samples to improve the classification performance of the model. The proposed SSMEKL can improve the performance of the classifier by using a small number of labeled samples and numerous unlabeled samples to improve the classification performance of MEKL. In the experiment, the results on four real-world data sets and two multiview data sets validate the effectiveness and superiority of the proposed SSMEKL.

The multi-label image classification problem is one of the most important problems in the field of computer vision, which needs to predict and output all the labels in an image. Multiple labels to be classified in an image increases the difficulty of image classification, and multi-label image classification usually requires additional attention to the positions of the object with different scales and poses. Hence, how to use the dependency relationship between labels to improve the recognition accuracy is an important problem when the object is difficult to directly identify. In this paper, we propose a designed network called the Semantic Supplementary Network with Prior Information (SSNP) to address this problem. The proposed SSNP first generates prior information by using a prior information network with different convolutional layers. Then the semantic supplementary module generates semantic information of the potential labels that is highly relevant to the current information based on the prior information, thereby effectively using the dependency relationship between the labels to improve the classification accuracy. Different from existing methods which pay more attention to the image feature extraction process, we focus on the impact of high-level semantic information generated after feature extraction on the results and tap the potential of high-level semantic information through a semantic supplementary module to strengthen the potential dependence between labels. Experimental results on public benchmark datasets demonstrate that the proposed architecture achieves the state-of-the-art performance, especially when predicting some semantically dependent labels.

The rich semantic information contained in images is an important clue to explore visual emotions. Therefore, exploring the correlation between visual emotion and the semantic relationship of objects, and extracting more effective semantic features through explicit or implicit modeling is very important for visual emotion analysis. In this paper, a novel Graph-based Object Semantic Refinement (GOSR) model is proposed to extract multi-level semantic features for visual emotion classification, in which graph structures is used to represent the object semantics and their position relationships of an image, and Graph Convolutional Networks (GCN) is used to refine object information by the aggregating neighbor object with their position relationships. The different convolutional layer’s features from GCN are further fused by Gated Recurrent Units (GRU) networks to achieve high-level semantic features. Then a framework with two branches to leverage visual and semantic information for visual sentiment analysis is proposed, which uses convolutional neural networks to extract visual features from images, and collaborates with semantic features from GOSR model to achieve better emotion recognition results. Besides, for alleviating the potentially unreasonable predictions and promote models collaboration, a novel tendency loss function based on the correlations among emotion labels is proposed to adjust the output activation value other than the target label. Extensive experiments on four widely used benchmark datasets show that our proposed method can achieve competitive performance and outperform most of the state-of-the-art methods on visual emotion recognition.

MOTIVATION: The discovery of cancer subtyping can help explore cancer pathogenesis, determine clinical actionability in treatment, and improve patients' survival rates. However, due to the diversity and complexity of multi-omics data, it is still challenging to develop integrated clustering algorithms for tumor molecular subtyping. RESULTS: We propose Subtype-GAN, a deep adversarial learning approach based on the multiple-input multiple-output neural network to model the complex omics data accurately. With the latent variables extracted from the neural network, Subtype-GAN uses consensus clustering and the Gaussian Mixture model to identify tumor samples' molecular subtypes. Compared with other state-of-the-art subtyping approaches, Subtype-GAN achieved outstanding performance on the benchmark datasets consisting of ∼4000 TCGA tumors from 10 types of cancer. We found that on the comparison dataset, the clustering scheme of Subtype-GAN is not always similar to that of the deep learning method AE but is identical to that of NEMO, MCCA, VAE and other excellent approaches. Finally, we applied Subtype-GAN to the BRCA dataset and automatically obtained the number of subtypes and the subtype labels of 1031 BRCA tumors. Through the detailed analysis, we found that the identified subtypes are clinically meaningful and show distinct patterns in the feature space, demonstrating the practicality of Subtype-GAN. AVAILABILITYAND IMPLEMENTATION: The source codes, the clustering results of Subtype-GAN across the benchmark datasets are available at https://github.com/haiyang1986/Subtype-GAN. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Sampling method is one of the most commonly used techniques in dealing with imbalanced data. Most of the existing undersampling methods randomly select samples from negative class with replacement. However, it may lose some important information of the training data. Moreover, increasing the positive data by oversampling in high imbalanced situations may cause the overlapping problem. To overcome these problems, this paper proposes a hybrid sampling method. The method takes the distributions of the training data into consideration by the information entropy, thus distinguishing the important samples in the undersampling procedure. Meanwhile, since the positive data only extend to the size of each subset of the negative class in the oversampling, the overlapping problem is relieved. Further, the method retains all the data in the training procedure and generates various data views from the original training data. Then each view is handled with an individual basic classifier. Finally, all the basic classifiers are combined by the ensemble method. The newly proposed method is named as Entropy-based Hybrid Sampling Ensemble Learning (EHSEL). In addition, the EHSEL is applied to three different kinds of basic classifiers to validate its robustness. Experiments results show the great effectiveness of the EHSEL on real-world imbalanced data sets.

Based on the current research on EEG emotion recognition, there are some limitations, such as hand-engineered features, redundant and meaningless signal frames and the loss of frame-to-frame correlation. In this paper, a novel deep learning framework is proposed, named the frame-level distilling neural network (FLDNet), for learning distilled features from the correlations of different frames. A layer named the frame gate is designed to integrate weighted semantic information on multiple frames to remove redundant and meaningless signal frames. A triple-net structure is introduced to distill the learned features net by net to replace the hand-engineered features with professional knowledge. Specifically, one neural network is normally trained for several epochs. Then, a second network of the same structure will be initialized again to learn the extracted features from the frame gate of the first neural network based on the output of the first net. Similarly, the third net improves the features based on the frame gate of the second network. To utilize the representation ability of the triple neural network, an ensemble layer is conducted to integrate the discriminative ability of the proposed framework for final decisions. Consequently, the proposed FLDNet provides an effective method for capturing the correlation between different frames and automatically learn distilled high-level features for emotion recognition. The experiments are carried out in a subject-independent emotion recognition task on public emotion datasets of DEAP and DREAMER benchmarks, which have demonstrated the effectiveness and robustness of the proposed FLDNet.

Given the complexity and diversity of the cancer genomics profiles, it is challenging to identify distinct clusters from different cancer types. Numerous analyses have been conducted for this propose. Still, the methods they used always do not directly support the high-dimensional omics data across the whole genome (Such as ATAC-seq profiles). In this study, based on the deep adversarial learning, we present an end-to-end approach ClusterATAC to leverage high-dimensional features and explore the classification results. On the ATAC-seq dataset and RNA-seq dataset, ClusterATAC has achieved excellent performance. Since ATAC-seq data plays a crucial role in the study of the effects of non-coding regions on the molecular classification of cancers, we explore the clustering solution obtained by ClusterATAC on the pan-cancer ATAC dataset. In this solution, more than 70% of the clustering are single-tumor-type-dominant, and the vast majority of the remaining clusters are associated with similar tumor types. We explore the representative non-coding loci and their linked genes of each cluster and verify some results by the literature search. These results suggest that a large number of non-coding loci affect the development and progression of cancer through its linked genes, which can potentially advance cancer diagnosis and therapy.

MOTIVATION: Given the complexity of genome regions, prioritize the functional effects of non-coding variants remains a challenge. Although several frameworks have been proposed for the evaluation of the functionality of non-coding variants, most of them used 'black boxes' methods that simplify the task as the pathogenicity/benign classification problem, which ignores the distinct regulatory mechanisms of variants and leads to less desirable performance. In this study, we developed DVAR, an unsupervised framework that leverage various biochemical and evolutionary evidence to distinguish the gene regulatory categories of variants and assess their comprehensive functional impact simultaneously. RESULTS: DVAR performed de novo pattern discovery in high-dimensional data and identified five regulatory clusters of non-coding variants. Leveraging the new insights into the multiple functional patterns, it measures both the between-class and the within-class functional implication of the variants to achieve accurate prioritization. Compared to other two-class learning methods, it showed improved performance in identification of clinically significant variants, fine-mapped GWAS variants, eQTLs and expression-modulating variants. Moreover, it has superior performance on disease causal variants verified by genome-editing (like CRISPR-Cas9), which could provide a pre-selection strategy for genome-editing technologies across the whole genome. Finally, evaluated in BioVU and UK Biobank, two large-scale DNA biobanks linked to complete electronic health records, DVAR demonstrated its effectiveness in prioritizing non-coding variants associated with medical phenotypes. AVAILABILITY AND IMPLEMENTATION: The C++ and Python source codes, the pre-computed DVAR-cluster labels and DVAR-scores across the whole genome are available at https://www.vumc.org/cgg/dvar. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.