Medical image datasets inherently suffer from class imbalance, label scarcity, and intra-class heterogeneity. Under semi-supervised learning, pseudo-label selection remains a central research focus. Class prototypes constructed in the latent space have been studied for pseudo-label selection, particularly because sub-class prototypes can effectively represent within-class heterogeneity in medical imaging datasets. However, since sub-class prototypes are constructed using a relatively small number of samples, they may lead to local optima during prototype updating and usage. In addition, loss weight rebalancing is commonly used to mitigate attention bias in model. Existing methods tend to focus exclusively on improving the minority class, without taking into account ambiguous samples, which are often difficult to learn and contain valuable information. Here, a prototype-driven framework PDMatch is proposed, incorporating Prototype-Driven Expert Voting (PDEV), Majority-class Pseudo-label Under-Sampling (MPUS) and Dual-Balanced Loss (DBL). First, an Expert Prototype Set is constructed, and hierarchical updating is proposed to sufficiently capture the feature distributions. PDEV then generates and filters pseudo-labels through prototype-driven voting, where multiple candidate sub-class prototypes are jointly considered. Second, MPUS offers a concise post-processing strategy that reduces imbalance in the pseudo-label distribution. Finally, Dual-Balanced Loss rebalances loss weights to jointly account for class distribution and sample difficulty. Comprehensive experiments are conducted on three datasets with distinct imbalance structures, demonstrating that our model addresses inherent data challenges and consistently outperforms state-of-the-art methods across multiple metrics. Additionally, we compare our model with clinical expert assessments to underscore its practical value in real-world applications. Source code is available at https://github.com/wang-su-lei/PDMatch .

BACKGROUND: Drug-drug interactions (DDIs) occur when two or more drugs are taken simultaneously or successively. Early detection of adverse drug interactions can be essential in preventing medical errors and reducing healthcare costs. Many computational methods already predict interactions between small molecule drugs (SMDs). As the number of biotechnology drugs (BioDs) increases, so makes the threat of interactions between SMDs and BioDs. However, few computational methods are available to predict their interactions. RESULTS: Considering the structural specificity and relational complexity of SMDs and BioDs, a novel multi-modal representation learning method called Multi-SBI is proposed to predict their interactions. First, multi-modal features are used to adequately represent the heterogeneous structure and complex relationships of SMDs and BioDs. Second, an undersampling method based on Positive-unlabeled learning (PU-sampling) is introduced to obtain negative samples with high confidence from the unlabeled data set. Finally, both learned representations of SMD and BioD are fed into DNN classifiers to predict their interaction events. In addition, we also conduct a retrospective analysis. CONCLUSIONS: Our proposed multi-modal representation learning method can extract drug features more comprehensively in heterogeneous drugs. In addition, PU-sampling can effectively reduce the noise in the sampling procedure. Our proposed method significantly outperforms other state-of-the-art drug interaction prediction methods. In a retrospective analysis of DrugBank 5.1.0, 14 out of the 20 predictions with the highest confidence were validated in the latest version of DrugBank 5.1.8, demonstrating that Multi-SBI is a valuable tool for predicting new drug interactions through effectively extracting and learning heterogeneous drug features.

In global climate researches, marine ecosystem researches, and ocean-related applications, it is of considerable significance to accurately observe and predict sea surface temperature (SST). However, various physical and environmental factors affect the changes in SST, making it highly random and uncertain. Therefore, it is still a challenge to propose a highly accurate SST prediction method. SST prediction methods based on the temporal information usually focus on capturing the temporal influence of the historical SST but ignore the spatial influence in the sea area, so these methods meet the performance bottlenecks. To fuse the multidimensional spatiotemporal influence and further improve the accuracy of the SST prediction, this letter proposed the convolutional gated recurrent unit (GRU) with multilayer perceptron (CGMP) to predict SST in the Bohai Sea and the South China Sea. The convolutional layer of CGMP can capture the neighbor influence effectively in the spatial dimension, making up for the shortcomings of methods that are based on the temporal information and do not consider the spatial information. The GRU layer and the MLP layer of CGMP can process historical information effectively in the temporal dimension. Experiments showed that the prediction performance of CGMP was better than those of other comparison methods in different sea areas, different schemas, and different prediction scales. Besides, the error distribution law in the Bohai Sea daily mean SST prediction was explored.

Identifying essential genes in comparison states (EGS) is vital to understanding cell differentiation, performing drug discovery, and identifying disease causes. Here, we present a machine learning method termed Prediction of Essential Genes in Comparison States (PreEGS). To capture the alteration of the network in comparison states, PreEGS extracts topological and gene expression features of each gene in a five-dimensional vector. PreEGS also recruits a positive sample expansion method to address the problem of unbalanced positive and negative samples, which is often encountered in practical applications. Different classifiers are applied to the simulated datasets, and the PreEGS based on the random forests model (PreEGSRF) was chosen for optimal performance. PreEGSRF was then compared with six other methods, including three machine learning methods, to predict EGS in a specific state. On real datasets with four gene regulatory networks, PreEGSRF predicted five essential genes related to leukemia and five enriched KEGG pathways. Four of the predicted essential genes and all predicted pathways were consistent with previous studies and highly correlated with leukemia. With high prediction accuracy and generalization ability, PreEGSRF is broadly applicable for the discovery of disease-causing genes, driver genes for cell fate decisions, and complex biomarkers of biological systems.

Carotid plaque classification plays a critical role in the identification of vulnerable plaques, so it is crucial for early risk estimation of cardiovascular and cerebrovascular events. Carotid ultrasound examination with ultrasound images and reports produced by professional doctors is the most common way to assess atherosclerotic plaques in clinical practice. However, existing deep learning methods for carotid ultrasound image analysis ignore the information in the ultrasound report. In this paper, we propose a multi-task learning (MTL) method named NDDR-LCS based on convolutional neural network (CNN) that leverages auxiliary information from ultrasound reports to assist the carotid plaque classification task. NDDR-LCS utilizes dense blocks as feature descriptors and organically combines three novel MTL mechanisms that are Neural Discriminative Dimensionality Reduction (NDDR), Learning Mixtures, and Cross-Stitch, to learn dependencies between ultrasound images and ultrasound reports. Based on carotid ultrasound images and their corresponding diagnostic reports, we conduct sufficient experiments to prove that NDDR-LCS outperforms state-of-the-art CNN methods for carotid plaque classification.

Differential network analysis has become an important approach in identifying driver genes in development and disease. However, most studies capture only local features of the underlying gene-regulatory network topology. These approaches are vulnerable to noise and other changes which mask driver-gene activity. Therefore, methods are urgently needed which can separate the impact of true regulatory elements from stochastic changes and downstream effects. We propose the differential network flow (DNF) method to identify key regulators of progression in development or disease. Given the network representation of consecutive biological states, DNF quantifies the essentiality of each node by differences in the distribution of network flow, which are capable of capturing comprehensive topological differences from local to global feature domains. DNF achieves more accurate driver-gene identification than other state-of-the-art methods when applied to four human datasets from The Cancer Genome Atlas and three single-cell RNA-seq datasets of murine neural and hematopoietic differentiation. Furthermore, we predict key regulators of crosstalk between separate networks underlying both neuronal differentiation and the progression of neurodegenerative disease, among which APP is predicted as a driver gene of neural stem cell differentiation. Our method is a new approach for quantifying the essentiality of genes across networks of different biological states.

Sea surface temperature (SST) prediction plays an important role in ocean-related fields. It is challenging due to the nonlinear temporal dynamics with changing complex factors and the inherent difficulties in long-scale predictions. Conventional models often lack efficient information extraction and cannot meet the requirements of long-scale predictions. Therefore, the gate recurrent unit (GRU) encoder-decoder with SST codes and dynamic influence link (DIL), GRU encoder-decoder (GED), which considered both the static and dynamic influence, is proposed in this letter. Each SST code, capturing the static information more effectively, was computed by all hidden states of the encoder and was individually associated with each predicted SST. The DIL, capturing the dynamic influence, connected the SST code with the early predicted future SST for solving the long-scale dependence problem. GED was tested on the Bohai Sea SST data sets and South China Sea SST data sets and compared with full-connected long-short term memory (FC-LSTM) and support vector regression. The results demonstrated that GED outperformed others on different prediction scales and different prediction terms (daily, weekly, and monthly), especially in terms of long-scale and long-term predictions. In addition, attention relationships between historical and future SSTs were further explored, and there was a meaningful finding that each future daily mean SST of Bohai Sea most strongly correlated with the past 27th to 29th historical values.

There are many algorithms for conducting alignments between undirected biomolecular networks (UBNs), including protein-protein interaction networks (PINs). However, none of them are specialized for directed biomolecular networks (DBNs), such as gene regulatory networks (GRNs) and metabolic networks (MNs). It is challenging and meaningful to achieve optimal mapping in DBN alignment. In this article, we propose a new algorithm, referred to as Directed Hungary Greedy Algorithm (DHGA), for the alignment of DBNs. DHGA focuses on a directed graph and catches information on the direction of edges. Furthermore, both the homology of biomolecules and the similarities of the network topologies are taken into consideration. In DHGA, expert knowledge can be brought in to pre-match biomolecules. We verified the effectiveness and robustness of DHGA using simulation datasets. Our experiments demonstrate that the performance of DHGA is clearly improved when expert knowledge is introduced. Moreover, we conducted DHGA on two metabolic pathway maps from KEGG and identified 21 pairs of similar cell cycle regulatory relationships between human and yeast, 12 of which were supported by references indicating that the paired relationships have the same function.

An efficient and reliable network alignment serves to find the mapping with maximum similarity between different networks, which is a way to improve our understanding of biology system and biological process. To build such network alignment is a foremost challenging task in computational biology because it involves the bipartite graph matching problem, which is an NP-hard problem. In this work, we propose a novel global biomolecular network alignment method named NFMA. Based on the minimum cost maximum flow problem (MCMFP) of network flow model, NFMA introduces opposite numbers to address the maximum cost problem so that the new model can get the alignment with maximum similarity. Furthermore, based on the existing method used to describe topological structure in undirected network, we present a new measure to identify topological information in directed network, which enables NFMA to extend to alignment in directed network. We conduct three experiments on classical yeast protein-protein interaction networks (PINs) and human PINs. We compare NFMA with other 7 eminent algorithms, and prove that NFMA can obtain an efficient and meaningful alignment which takes both biological similarity and topological similarity into consideration, and the time efficiency of NFMA is also outstanding in that it only takes approximately 250 seconds to get an accurate alignment when aligning two large networks.

Cardiovascular disease (CVD) caused by atherosclerosis is one of the major causes of death world-wide. Currently, diverse machine learning models have been applied to disease prediction and classification. However, most of them tend to focus on the performance of the algorithm and neglect the underlying variables for patients in different carotid atherosclerotic stages. In this paper, we propose a novel hybrid machine learning method named Subset Learning (S-learning) to predict and discover the risk factors of these different stages. The S-learning algorithm can elucidate the variables that have significant influence on the outcome of carotid atherosclerotic. Performance comparisons are based on the dataset collected from both Shanghai Renji and Shanghai Huashan Hospital. The result shows that the proposed method has superior classification performance than other classification algorithms. Our findings point to the utility of predictive machine learning and the discovery of risk factors to refine the treatment plans.

An agent is introduced into workflow management of PSE-Bio, and the agent composition model for workflow in Web service-based PSE is also proposed in this paper. In this model, workflow can cooperate with three kinds of autonomy agent which are interface agent, task agent and resource agent, so as to realize heterogeneous services integration and distributed task management based on intelligence and negotiation mechanism of agent. With this model, resource encapsulation and monitoring of service status can be implemented, the flexibility of workflow execution mode can be enhanced, and the self-determination of user to interact with workflow is improved. The workflow can be more adaptable to bioinformatics PSE in grid environment. The function modules, categorization and schedule strategy in this model are discussed. This model is implemented with protein interaction network search as an example.

To utilize technology of various areas effectively and efficiently, a grid enabled problem solving environment for bioinformatics (PSE-Bio) is built. This paper describes the architecture of PSE-Bio and proposes the new layer, Agent bundle, based on classical architecture with three layers. As a service bundle, Agent provides common single mechanism to access grid services, by which portal services can communicate with key grid components. Agent bundle not only search and select appropriate software in grid, but also have strategies to reuse local legacy program to save time and cost. Users can utilize PSE-Bio easily without any details about grid technology because PSE-Bio is service- oriented. An example application of PSE-Bio workflow is implemented.