The properties and performance of copolymers are fundamentally governed by their composition and molecular weight; however, achieving tailor-made polymer synthesis via conventional free radical polymerization remains a long-standing challenge. To address this challenge, machine learning is directly embedded into the polymerization workflow to establish a multifunctional polymerization platform that maps reaction parameters to polymer structures in silico. The platform achieves high predictive accuracy for molecular weight, copolymer composition, and conversion, while accurately predicting the composition drift with the conversion. Moreover, multivariate feature analysis reveals that the concentrations of initiator/co-initiator, monomer, and chain transfer agent, together with the initial monomer feed ratio, are the dominant factors governing experimental outcomes. Notably, the platform enables inverse design by prescribing experimental conditions for predesigned polymer, allowing the experimental synthesis of copolymers with predefined compositions and molecular weights with negligible relative errors. This study provides a proof-of-concept that machine learning-assisted conventional free radical polymerization enables on-demand polymer synthesis, thereby offering a promising pathway to transform polymer development from empirical trial-and-error to data-driven fabrication.

Polypharmacy has become essential in managing complex and chronic conditions, yet it introduces significant clinical risks due to potential drug-drug interactions (DDIs). Existing computational models struggle to provide accurate and interpretable predictions, largely due to fragmented integration of molecular structures and biomedical knowledge. These limitations arise from static graph designs, inadequate substructure modeling, and insufficient incorporation of semantic context, highlighting the need for a unified, explainable solution. In this study, we propose MKGFlow-DDI, a multi-view knowledge-guided framework that jointly leverages drug-drug interaction networks and biomedical knowledge graphs to dynamically construct drug-flow subgraphs. The model incorporates a dual-channel encoder designed to capture atom-level information and substructure-level features, integrating them with the global semantic embeddings of the composite network to derive novel feature representations. These representations are fused to initialize node features within each subgraph, which are iteratively optimized through similarity-based edge refinement to reduce noise and enhance biological relevance. To further improve generalization and stability, a contrastive learning module is introduced to align representations of perturbed subgraphs by maximizing consistency across positive and negative sample pairs. Experimental results on DrugBank and TWOSIDES demonstrate that MKGFlow-DDI outperforms state-of-the-art baselines, especially in scenarios involving previously unseen drugs. Additionally, the model produces interpretable semantic pathways that align with known pharmacological mechanisms, enabling clinically meaningful insights. Overall, MKGFlow-DDI establishes a robust and biologically grounded approach to DDI prediction, offering a promising direction for computational pharmacovigilance and personalized therapy optimization.

Identifying drug–target interactions (DTIs) is essential in drug discovery and repositioning. Recently, deep learning has become the mainstream methodology for DTI prediction. However, the scarcity of three-dimensional structural data has forced almost all methods to predict drug-target interactions with low-dimensional data, thereby constraining their overall performance. In tackling this challenge, we introduce a novel approach, CLSF-DTI. CLSF-DTI incorporates high-dimensional structural and functional information into the drug and protein features through contrastive learning during the feature extraction stage. This ensures that the model no longer solely focuses on sequence information, leading to a more precise modeling outcome. Experiments on five benchmark datasets demonstrate that CLSF-DTI achieves the best overall performance among five state-of-the-art baselines. Through ablation studies, we further prove that the contrastive module enhances the predictive performance and generalization ability of CLSF-DTI. Moreover, CLSF-DTI successfully identified some ligands for the protein PKA-C α in drug screening experiments. This study proposed a contrastive learning model CLSF-DTI that integrates structural and functional similarity. It outperforms existing methods in drug-target interaction prediction and has stronger generalization ability. However, the handling of unbalanced data and long-distance dependencies still needs to be improved in the future. The data and source code are available at https://github.com/ZhangLab312/CLSF_DTI.

Polymers, as foundational materials in modern industry, face persistent challenges in precision design and performance improvement due to structural intricacy, multifunctionality requirements, and sustainability imperatives. Machine learning (ML) has emerged as a transformative tool for elucidating structure–property correlations and expediting polymer material discovery. This review systematically examines ML applications across three domains: autonomous synthesis via reaction kinetic modeling, cross-scale property prediction linking polymeric configurations to bulk behavior, and sustainability-driven design frameworks. For automation synthesis, ML integrates polymerization kinetics with structure control and polymerization efficiency, enabling closed-loop systems for autonomous process refinement. In performance prediction, ML deciphers hierarchical architectures relationships with thermal resilience, optoelectronic responses, and mechanical robustness, providing physicochemical theory frameworks for tailored material design. Critical analyses address persistent limitations, including data paucity in specialty polymer classes, interpretability deficits in multimodal architectures, and validation gaps between simulation and experiments. By synergizing generative algorithms with high throughput experimentation, this strategy transcends empirical trial-and-error approaches, establishing a computational design paradigm spanning molecular-to-bulk scales. The resultant synergy between computational intelligence and polymer science not only streamlines material discovery cycles but also unlocks sustainable solutions for energy storage, eco-friendly materials, and adaptive smart systems, heralding a new era of data-driven macromolecular engineering.

Single-cell annotation, which identifies specific cell types in biological tissues, is a cornerstone of targeted therapy in precision medicine. However, the lack of incremental annotation capabilities in existing methods limits their adaptability to dynamic immune environment changes and broader applications. Additionally, the high-dimensional sparsity and long-tail distribution of single-cell data hinder practical long-tail incremental annotation. This paper introduces a single-cell incremental annotation framework that combines a distribution-aware diffusion model with an expression-aware knowledge distillation architecture to address these challenges. The distribution-aware module retrospects and replays historical data distributions, enabling precise generation of previously annotated cell types and mitigating catastrophic forgetting. The expression-aware module aligns gene representations through a multi-perspective attention mechanism, enhancing sensitivity to novel cell types while retaining knowledge of dominant categories. A fuzzy incremental guidance mechanism with uncertainty constraints further reduces the adverse effects of long-tail distributions, ensuring more robust annotation perfor mance. Experimental results validate the effectiveness of this framework, demonstrating significant improvements in multi-session incremental annotation accuracy. In addition, this framework enhances the biological and clini cal relevance of incremental annotation, enabling continuous integration of new datasets for improved discovery of rare cell types and disease progression analysis. This approach offers a scalable and generalizable solution to the challenges of long-tail incremental annotation in single-cell data.

Single-cell RNA sequencing (scRNA-seq) is entering an era of foundation models that accept the complete gene atlas as input, yet most current datasets cover only 10-12 k genes and contain numerous technical zeros, severely limiting the generalization of these models in downstream tasks. To address this, we pioneer the gene-completion task for scRNA-seq and present SAD, a diffusion-based framework tailored to extremely sparse data, capable of completing genes and correcting sparsity bias under high missing rates. Unlike imputation or reconstruction methods that rely on the i.i.d. assumption, SAD's completion paradigm can generate gene entries originally absent from the expression profile, be aware of and rectify sparsity-distribution bias, and supply foundation models with consistent, reliable inputs of more than 30 k genes. Extensive benchmarks show that SAD significantly outperforms existing methods across multiple completion metrics, particularly in extreme scenarios with missing rates above 80%. This provides a data foundation for reusing missing scRNA-seq information and for precision-medicine applications.

Accurate prediction of drug-drug interactions (DDIs) is critical for ensuring patient safety in polypharmacy, yet remains challenging due to the complexity of the underlying biochemical mechanisms. Existing methods are limited by inadequate fusion of heterogeneous features, insufficient integration of local and global molecular characteristics, and lack of interpretability for novel drugs. These limitations stem from fragmented processing of structural and biological data, which overlooks hierarchical feature relationships and latent pharmacological associations. Here, we present MCFusion-DDI, a multimodal framework that unifies molecular substructures, chemical fingerprints, and drug similarity networks via three synergistic encoding channels. Our approach leverages a graph neural network with bond-level attention for hierarchical substructure extraction, a convolutional network for global fingerprint encoding, and similarity integration through network embeddings. The core innovation is a dual-channel cross-attention mechanism that dynamically integrates intra-drug feature dependencies and inter-drug interactions using efficient linearized attention combined with gated fusion. Predictions are generated using a Kolmogorov-Arnold network with adaptive activation functions. Extensive evaluations demonstrate state-of-the-art performance, with MCFusion-DDI achieving 8.2 - 12.7% improvements in AUROC and F1 score across the DrugBank, ZhangDDI, and ChCh-Miner datasets, exhibiting exceptional robustness in novel-drug scenarios. Attention visualizations provide mechanistic interpretability by highlighting high-risk substructures such as competitive hydroxyl groups in nonsteroidal anti-inflammatory drugs. This work establishes a new paradigm for explainable DDI prediction in precision medicine.

Accurate characterization of cell-specific drug responses is a prerequisite for linking molecular mechanisms to patient-level therapeutic variability. Heterogeneous cellular responses substantially complicate the inference of drug effects from single-cell transcriptomic data. To address this, we introduce scDPR, the single-cell Drug Perturbation Responses framework. This framework predicts drug-perturbed single-cell transcriptomic states and decomposes observed responses into distinct causal effects. The framework consists of two modules: an attribute adaptation module that models drug-induced transcriptional shifts at the single-cell level; and a causal graph learning module that combining optimal transport (OT) infers direct drug effects while accounting for confounding influences. By integrating drug molecular features, dosage information, and cell-specific attributes, the model learns a latent representation that captures cell-specific transcriptional responses to drug perturbations. We conducted systematic evaluations on large-scale single-cell perturbation datasets, including L1000 and sci-Plex3. Experimental results demonstrate that scDPR outperforms state-of-the-art methods such as chemCPA, in forecasting transcriptome responses to unseen compounds and unknown pathways. It also offers insights into the cellular heterogeneity of drug responses, identifying key subpopulations that contribute to variability in treatment outcomes. In addition, compared to mainstream basic models, scDPR achieves better performance with shorter training time and fewer rounds, providing efficient computational support for large-scale drug screening and mechanism research.

Anticancer peptides (ACPs) exhibit significant pharmacological potential in studying cancer and other diseases. Although anticancer peptides have been extensively investigated, many remain undiscovered and hold promise as future therapeutics. Computational methods offer efficient solutions for identifying novel anticancer peptides, but current approaches often rely heavily on local amino acid sequence information, overlooking global interactions, which limits their performance and generalizability. Here, we present an explainable multiscale representation learning framework that integrates sequence information from bidirectional encoder and handcrafted features to enhance anticancer peptides predictions. Our model learns sequence representations at two scales: it captures local representations through handcrafted feature methods at residue scale and global representations through a bidirectional encoder at protein scale. In addition, it employs a cross-attention mechanism to fuse these two types of representations automatically. Compared to state-of-the-art techniques, our model demonstrates superior performance with a significant improvement of 2.3%, 1.3%, 2.5% and 3.2% in prediction accuracy. To further demonstrate the superiority of our model, it can achieve optimal performance even under conditions where negative samples are randomly generated in the alternate dataset and other datasets, resulting in a significant improvement in accuracy over current state-of-the-art techniques. Moreover, model provides “white-box” prediction, revealing representation shifts during inference and effectively identifying important subregions of sequences, thereby uncovering sequence motifs. Shapley additive explanation value analysis was performed on the fused features, revealing that certain features contribute significantly to the model’s predictions. In general, our model is a powerful tool for advancing artificial intelligence-driven drug discovery.

Cell type annotation is paramount for subsequent biomedical research endeavors. Recently, supervised classification methods based on deep learning have demonstrated significant advancements in cell type annotation. However, challenges persist regarding how to efficiently leverage single-cell multi-omics data for cell type annotation and interpretation and how to endow models with adaptive generalization capabilities for efficient identification of rare cell types. We introduce scMoAnno, a methodology for cell type annotation based on single-cell multi-omics data. Leveraging pre-trained cross-attention networks, scMoAnno facilitates efficient mutual learning and fusion of genetic and epigenetic distributions in single-cell multi-omics data, enabling precise cell type annotation. Predictions conducted by scMoAnno on single-cell RNA sequencing (scRNA-seq) and single-cell assay for transposase-accessible chromatin sequencing (scATAC-seq) data demonstrate enhanced generalization capacity for the recognition of rare cell types through pre-trained feature fusion extraction and training of classifiers based on fused data inputs. Rigorously conducted experiments on four meticulously curated benchmark datasets illustrate that scMoAnno surpasses nine other state-of-the-art baseline models in overall performance. Furthermore, the increase in clustering scores, fused with multi-omics data and extracted by scMoAnno, is elucidated from the perspective of promoting downstream analysis of single cells, explaining how our model's enhancement of cell type annotation performance yields positive effects. The source code and data are publicly available at https://github.com/ZhangLab312/scMoAnno.

Single-cell type annotation is crucial for understanding the tumor microenvironment. With advances in single-cell sequencing, researchers can now analyze the functional characteristics of different cell types at the single-cell level. However, high-dimensional sparsity and technical noise in single-cell data pose challenges for existing annotation methods. Traditional approaches rely mainly on transcriptomic data, neglecting complementary multi-omics information, leading to information loss and reduced accuracy. To address these limitations, this study proposes a novel single-cell annotation method based on multimodal alignment. By designing two alignment strategies—inter-modal feature alignment and inter-model knowledge alignment—the method integrates scRNA-seq and scATAC-seq data. The inter-modal feature alignment strategy corrects distributional discrepancies between modalities, enabling seamless mapping in the latent space and more accurately capturing multi-omics characteristics. The inter-model knowledge alignment strategy transfers structured knowledge from a teacher model to a student model, balancing computational efficiency and annotation performance. This approach reduces reliance on single modalities in traditional methods and enhances adaptability to diverse experimental conditions. Experiments on public datasets show that the proposed method outperforms state-of-the-art single-omics models across multiple benchmarks, demonstrating strong annotation capabilities and superior generalization to rare cell types.

The Gene Regulatory Networks (GRNs) ensure the stability of cellular states, preserving specific phenotypes and functions throughout the differentiation process. However, current tools still need improvement to effectively integrate multi-omics data and infer GRNs for particular cell types. We introduce CLMOGRI, a multi-omics TF-gene regulatory network inference framework based on heterogeneous networks and contrastive learning, designed to integrate multi-omics data for GRN inference. Through random walk techniques, CLMOGRI embeds multi-omics data into a unified feature space and extracts similar features between nodes. It then measures node similarity and predicts node relationships by contrastive learning. Finally, it includes a regulatory network interpreter to identify critical nodes and modules in GRNs, offering an analytical method for understanding complex interactions within biological systems. CLMOGRI surpasses existing baseline methods in terms of Area Under the Precision-Recall Curve (AUPR) and F-Score metrics, indicating its efficacy in capturing multi-omics information for GRN inference. It also reveals vital nodes and modules within the gene regulatory network, improving the interpretability of CLMOGRI and the utility of GRNs.

Drug-target interaction (DTI) prediction is a critical step in drug discovery. Despite significant advances in deep learning-based methods, data representation and feature alignment challenges remain. Specifically, previous methods often rely on low-dimensional representations and limited labeled data, overlooking the importance of high-dimensional spatial geometric information and unlabeled data, which restricts the extraction of crucial features. Additionally, most approaches align features between amino acid residues and drug atoms using dot-product similarity, ignoring their biochemical differences, which limits the effectiveness of the alignment. We propose GGANet, a geometry-enhanced gated attention network, to address these limitations. GGANet integrates low-dimensional pre-trained embeddings with high-dimensional geometric information to generate robust and generalizable feature representations while employing gated attention to ensure efficient feature alignment. We conducted experiments on four datasets, and the results show that GGANet outperforms baseline methods, especially in predicting unseen data, demonstrating superior robustness and generalization. The implementation details of GGANet are available at https://github.com/ZhangLab312/GGANet.

Accurately predicting individual cancer drug response(CDR) remains a significant challenge. The accumulation of multi-omics data from cell lines and drug information has greatly facilitated the development of predictive models for CDR. However, effectively integrating multi-source data and improving the prediction accuracy of models remain critical challenges to be addressed. In this study, based on the Genomics of Drug Sensitivity in Cancer (GDSC) database, designed a CDR prediction model named AGCCK(Autoencoder-Graph Neural Networks-Convolutional Neural Networks-Cross-Attention-Kolmogorov-Arnold Network), which utilizes Graph Neural Networks (GNN) and Cross-Attention mechanisms. For multisource data, the model incorporates different feature extractors, employing GNN or Convolutional Neural Networks (CNN) to extract features. After extraction, the cell line and drug data are fused using a cross-attention module, followed by another cross-attention module to further integrate the cell line and drug features. Finally, a Kolmogorov-Arnold Network (KAN) module is used to predict the final outcome. Research indicates that this is the first attempt to combine cross-attention modules and KAN in the field of CDR prediction. A series of model evaluation experiments were conducted on the GDSC database. The results demonstrate that the model’s predictive performance surpasses existing state-of-the-art algorithms, with improvements of 1.8%, 2%, and 2.8% in the Pearson Correlation Coefficient (PCC), Spearman Correlation Coefficient (SCC), and Coefficient of Determination (R2), respectively, compared to the best baseline model. The model also exhibits strong predictive capabilities in missing value prediction tasks. This study holds significant importance for CDR prediction and new drug development. AGCCK is freely available at https://github.com/ZhangLab312/AGCCK.

Precision prediction of cancer drug response remains a critical challenge in personalized medicine. With ongoing advancements in related research, substantial amounts of cell line omics data and drug feature information have been accumulated, offering valuable resources for investigating cancer drug responses. However, effectively integrating these multi-omics features and constructing accurate and interpretable network-based prediction models remain challenging. To address these issues, we propose GraphTCDR, a model based on heterogeneous graph neural networks and multi-omics data that can accurately predict cancer drug responses. The specific workflow of GraphTCDR is as follows: First, a cell line-drug heterogeneous network is constructed, using multi-omics data and drug features as node attributes. Next, node feature learning is conducted on the heterogeneous network. Finally, the learned features are fed into fully connected layers to predict IC50 values. Extensive experiments on the PRISM database demonstrate that GraphTCDR outperforms existing state-of-the-art methods across all evaluation metrics. Compared with the current best-performing model, GraphTCDR achieves improvements of 3.60 % in PCC, 4.30 % in SCC, 6.50 % in R2, and a 1.60 % reduction in RMSE. The reliability of GraphTCDR's predictions on unlabeled samples is also validated. Moreover, GraphTCDR maintains stable performance even when the amount of training data is reduced, unlike other algorithms, indicating superior robustness. GraphTCDR offers a novel approach to drug response prediction and has significant implications for advancing personalized cancer therapy.

Multiomics integration of single-cell datasets generated from multiple omics technologies is crucial for defining cellular heterogeneity. Mosaic data contains any combination of different modalities and batches of data, which brings challenges such as incomplete reference samples, non-overlapping features, and batch effects, making it difficult to extract information from them. Most existing methods require a reference batch that contains all modalities, which limits the versatility of mosaic data. In this study, we proposed a single-cell mosaic data integration method that guides variational cross-modal fusion via regulation graphs (GVMOS). Based on mosaic data, GVMOS first generates a regulatory graph of cross-modal feature interactions. Then, a coupled-trained dual variational autoencoder module is used to extract the complementary and specific information in different modal data and the cross-modal feature topology information in the regulatory graph. By integrating modality information with feature information, GVMOS can guide modality integration from the perspective of spatial semantics. Experiments on three bimodal paired datasets and five bimodal mosaic combinations show that GVMOS significantly improves the integration performance of existing methods, with an average improvement of 3% in overall scores.

The rapid accumulation of single-cell RNA sequencing data has provided unprecedented computational resources for cell type annotation, significantly advancing our understanding of cellular heterogeneity. Leveraging gene expression profiles derived from transcriptomic data, researchers can accurately infer cell types, sparking the development of numerous innovative annotation methods. These methods utilize a range of strategies, including marker genes, correlation-based matching, and supervised learning, to classify cell types. In this review, we systematically examine these annotation approaches based on transcriptomics-specific gene expression profiles and provide a comprehensive comparison and categorization of these methods. Furthermore, we focus on the main challenges in the annotation process, especially the long-tail distribution problem arising from data imbalance in rare cell types. We discuss the potential of deep learning techniques to address these issues and enhance model capability in recognizing novel cell types within an open-world framework.

Background and Objective: Enhancer activity is central to the regulation of gene expression. Despite extensive computational research on enhancers, there is still a lack of understanding of the relationship between DNA sequences and enhancer activity, overlooking the importance of sequence scale, which poses particular challenges for predicting enhancer activity. This study aims to develop an interpretable framework for predicting enhancer activity by: (a) using longer-scale information and explore the relationship between scale and task sensitivity; and (b) designing multi-angle experiments to analyze key motifs and conduct cross-cell validation. Methods and Results: This research have constructed LEAP (Learn Enhancer Activity and Predict), an interpretable framework based on the Performer model. It directly captures longer-scale information on enhancers from DNA sequences for predicting enhancer activity and exploring enhancer grammar. Compared to previous models with shorter inputs, this architecture first integrates gene information within a 1001bp range of enhancers to achieve accurate activity prediction and conducts exploration at multiple scales. Experiments show that LEAP outperforms the currently published state-of-the-art(SOTA)methods, achieving the highest PCC and the lowest MSE,the mean values reached 0.794 and 0.406 across all datasets, respectively.Contribution score visualization delineates the heterogeneous functional importance across critical motifs. Conclusions: Our study underscores that LEAP provides a new perspective for predicting and analyzing enhancer activity.it explores the impact of DNA sequences of different bp lengths on enhancer activity prediction, analyzing motifs that have a key influence on the prediction task from multiple perspectives, providing insights into the internal grammar of enhancers, thereby better understanding the significance of long-sequence enhancers.

Electroencephalogram (EEG) signals, as a reliable biological indicator, have been widely used in fatigue driving detection due to their capacity to reflect a driver's cognitive and neural response state. However, EEG signals have problems such as imbalanced data distribution, significant differences between subjects, and complex scenes, which affect the detection effect. Small commonalities between input objects can be interpreted as important information about an entire sample. Therefore, to retain as much information as possible, We design a new approach for integrating fuzzy features, comprehensive adaptive interpretable TSK fuzzy classifier(CAI-TSK-FC). It not only captures the features of multiple subclassifiers more efficiently and alleviates the dataset imbalance problem. Also, it can reduce the accumulation of error information by randomly retaining fuzzy rules as well as normalization. Finally, we linearly combine the results of multiple subclassifiers to comprehensively consider the learning effect of multiple subclassifiers to adapt to different subjects and datasets. Experiments conducted on both self-made and public datasets (SEED-VIG) show that CAI-TSK-FC has good performance and interpretability on different EEG fatigue driving datasets. In comparison to existing methods, it achieves an accuracy improvement of 3.15% and 1.52%, respectively, as well as a specificity improvement of 4.72% and 0.91%, respectively.