Token aggregation converts token-level representations into fixed-dimensional sample representations, but most pooling methods operate only in the original token space. We introduce Frequency-Domain Latent-attention Gated Pooling (FLaG), a plug-in aggregation module that re-expresses encoder outputs in the Fourier domain before final pooling. FLaG represents the nonredundant rFFT spectrum through concatenated real and imaginary components, summarizes spectral tokens with learnable latent queries, derives a sample-conditioned channel gate, and reconstructs modulated token representations for downstream aggregation. We evaluate the same architecture across ESM2-based antimicrobial peptide (AMP) activity prediction, ResNet18 image classification on CIFAR-10 and CIFAR-100, and three RoBERTa-based language tasks. FLaG achieves the best macro-averaged Spearman correlation coefficient, RMSE, and Recall@50 across four AMP backbone-species settings and the highest top-1 accuracy on CIFAR 10. It also achieves the best mean results on five of seven language metrics, although mean pooling remains strongest on STSBenchmark. AMP-side mechanistic analyses reveal low-frequency prediction sensitivity across most encoder layers, with increased relative high-frequency sensitivity in the final layer, and pronounced peptide-specific positional responses. The residual gate broadly amplifies spectral channels while preserving the low-frequency-dominated energy profile, whereas latent cross-attention exhibits sample- and species-specific spectral allocation. Overall, FLaG provides a transferable frequency-domain aggregation bias across protein, visual, and textual representations, with benefits that depend on the backbone and downstream task. Supplementary materials, source code, and data are available at https://www.healthinformaticslab.org/supp/ and https://github.com/Kewei2023/AMPCliff/tree/FLaG.

The widespread adoption of electronic medical records (EMR) has facilitated the prediction of patient prognosis and disease progression, yet inherent issues such as irregular sampling and missing values continue to pose challenges for clinical time-series analysis. This study aims to develop a robust framework capable of effectively handling incomplete EMR data while capturing complex temporal patterns and feature interaction. We propose MedGAITS, a novel two-stage graph autoencoder framework for irregular and incomplete clinical time series. The model employs a progressive learning strategy: the first stage performs coarse-grained reconstruction of the zero-filled input via dynamic graph learning, while the second stage refines this representation to extract deep, robust features. Through iterative dynamic graph construction and residual-style information propagation, MedGAITS learns uncertainty-aware representations directly from raw, partially observed data, avoiding the biases introduced by explicit imputation. MedGAITS achieved competitive or superior performance compared to state-of-the-art models across multiple public datasets (PhysioNet 2012, COVID-19, and eICU) in both regression and classification tasks. Meanwhile, MedGAITS provides clinically interpretable insights into COVID-19 progression, identifying neutrophils and LDH as early biomarkers and white blood cell count as a later-stage indicator, thereby characterizing the disease’s temporal profile. MedGAITS provides an effective solution for handling irregular clinical time-series data with missing values. Its two-stage imputation-and-representation learning design not only improves performance in downstream predictive tasks but also helps uncover clinically meaningful, time-evolving features, offering valuable insights for disease monitoring and biomarker discovery.

Activity-cliff ranking remains difficult because local structural changes can cause large activity differences, while high-quality data that resolve the underlying mechanisms remain limited. To use available activity labels more effectively, we combine absolute-activity regression with ranking-consistency learning. CliffRank trains two parallel predictors with mean squared error, a thresholded listwise loss, and Pairwise Preference Consistency (PPC), which aligns relative ordering in the preference-probability space. On three antimicrobial peptide datasets, CliffRank with ESM2-t12 achieved the highest mean Spearman correlation of 0.5393 and mean Recall@50 of 21.4, although the leading method varied across individual datasets. On three small-molecule datasets, CliffRank with PNA, where PPC was activated after 120 epochs, achieved the highest mean Spearman correlation of 0.6890, while its mean Recall@50 of 30.4 matched that of ACANet-PNA. The PPC results also define its practical limits. Asymmetric initialization improved the MolCLR-GIN averages but did not improve every target. For PNA without pretrained weights, delayed PPC improved selected metrics, but no schedule was best for both mean Spearman correlation and mean Recall@50. Future work should evaluate more targets and antimicrobial peptide systems, develop adaptive PPC schedules, and incorporate protein or membrane context when available.

Competitive endogenous RNA (ceRNA) networks are pivotal for uncovering disease molecular mechanisms. Graph representation learning is a cornerstone for modeling biological regulatory networks and predicting disease-related biomarkers. However, current methods face challenges: traditional graph neural network (GNN) rely on low-order graph structures, which struggle to capture highorder molecular interactions, resulting in topological information loss; shallow GNN fail to model long-range dependencies, while deep architectures suffer from oversmoothing, limiting complex regulatory expression; static embeddings overlook dynamic molecular interactions, reducing biomarker accuracy. These limitations highlight the need for advanced graph learning frameworks. To address these challenges, we propose DMHLF, a Dynamic Multi-scale Hypergraph Learning Framework for predicting disease-associated ceRNA biomarkers. The framework first integrates multiple regulatory relationships among miRNAs, lncRNAs, circRNAs, mRNAs, and diseases to construct disease-specific ceRNA regulatory networks, capturing local and global regulatory patterns through multi-Hop hyperedges. Subsequently, we devise a HypergraphWeighted Dynamic Random Walk (HEDRW) method to dynamically extract node meta-embeddings that encode high-order regulatory information. Concurrently, we extend Eigen-GNN spectral analysis to hypergraph structures, incorporating a residual-enhanced hypergraph neural network to preserve the global topological properties of shallow hypergraphs. Finally, a cross-scale attention mechanism aligns and fuses multi-scale features to generate high-quality node embeddings for disease-ceRNA association prediction. Experiments on diverse datasets demonstrate that DMHLF significantly outperforms existing methods. Case study further validates the framework's efficacy in identifying disease-related ceRNA biomarkers, providing a reliable predictive tool for biomedical research.

The rapid accumulation of Electronic Health Records (EHRs) has transformed healthcare by providing valuable data that enhance clinical predictions and diagnoses. While conventional machine learning models have proven effective, they often lack robust representation learning and depend heavily on expert-crafted features. Although deep learning offers powerful solutions, it is often criticized for its lack of interpretability. To address these challenges, we propose DeepSelective, a novel end to end deep learning framework for predicting patient prognosis using EHR data, with a strong emphasis on enhancing model interpretability. DeepSelective combines data compression techniques with an innovative feature selection approach, integrating custom-designed modules that work together to improve both accuracy and interpretability. Our experiments demonstrate that DeepSelective not only enhances predictive accuracy but also significantly improves interpretability, making it a valuable tool for clinical decision-making. The source code is freely available at http://www.healthinformaticslab.org/supp/resources.php .

Token aggregation is a common bottleneck in models that map token representations to sample-level predictions, yet most pooling methods operate only in the original token domain. We propose FLaG, a plug-in aggregation module that transforms token representations with the real FFT, summarizes spectral components with learnable latent queries, applies a channel-wise gate, and reconstructs enhanced time-domain tokens for final pooling. We evaluate FLaG on antimicrobial peptide (AMP) activity prediction with ESM2, image classification with ResNet18 on CIFAR-10 and CIFAR-100, and text classification with RoBERTa on IMDB and GLUE. FLaG achieves its clearest gains on the ESM2-8M antimicrobial peptide tasks and on CIFAR-100, while remaining competitive with strong text baselines on IMDB and GLUE. Then we probe its behavior on the AMP setting with band knockouts, gate summaries, residue perturbations, latent-query readouts, and structure-proxy stratification. We find that low-frequency bands contribute the most overall, and the remaining higher-band pattern is more sample-specific. The gate acts as a broadly shared spectral reweighting stage and the cross-attention patterns are sample-specific with mild query-wise differentiation, and higher-helix peptides exhibit stronger average spectral sensitivity in both bacteria. The supplementary materials, source code and data are released at https://www.healthinformaticslab.org/supp/ and https://github.com/Kewei2023/AMPCliff/tree/FLaG.

Introduction Feature selection plays a crucial role in improving predictive performance and interpretability in high-dimensional machine learning tasks. However, it is an NP-hard combinatorial optimization problem. Conventional heuristic or greedy approaches are prone to premature convergence and often yield locally optimal feature subsets, especially in high-dimensional settings with complex feature interactions. Objectives This study aims to enhance existing feature selection methods by identifying complementary helper features overlooked by conventional approaches through a plug-and-play refinement framework. Methods We formulate Conditional Feature Selection as augmenting an existing feature subset by discovering complementary helper features from the residual feature space, i.e., features typically discarded by conventional selection methods. Helper-Enhanced Feature Selection (HeFS) implements this idea as a post-selection refinement framework that can be integrated with diverse baseline algorithms. By incorporating biased initialization and ratio-guided mutation and employing a multi-objective search strategy that balances predictive accuracy and feature complementarity, HeFS enables the systematic identification of informative yet previously overlooked features. Results A comprehensive evaluation on 18 benchmark datasets shows that HeFS consistently enhances baseline feature selection performance by identifying informative helper features overlooked by conventional methods. On average, the helper features identified by HeFS improve classification accuracy by more than 9% over traditional feature selection methods and 6% over state-of-the-art approaches, with the largest gain observed on the Spambase dataset, where accuracy increases by 31.8% over the MGWO baseline. Further analysis on molecular datasets using Pearson correlation coefficients indicates that helper features exhibit substantially lower correlation with baseline-selected features (0.06) than the correlations among baseline features themselves (0.24), suggesting reduced redundancy and complementary contributions. Conclusion By explicitly modeling feature complementarity in the residual feature space, HeFS mitigates the limitations of conventional heuristic or greedy feature selection methods and improves classification performance across diverse application domains.

Ultrasound is the primary imaging modality for assessing thyroid nodules, and the ACR TI-RADS framework standardizes diagnosis through five ultrasound feature categories that are aggregated into five risk levels (TR1-TR5). Although widely adopted in clinical practice, most deep learning approaches focus on binary malignancy classification, while multi-class prediction and explicit utilization of feature-level supervision remain underexplored, largely due to limited annotated data. In this study, we introduce the STN dataset of 600 thyroid nodules with paired transverse and longitudinal ultrasound images, bounding box annotations, and complete labels for all five TI-RADS feature categories. Following the clinical decision process, we investigate how structured feature information can guide representation learning during training while requiring only images at inference. We demonstrate that text embeddings derived from standardized feature descriptions form a stable surrogate representation for TI-RADS risk levels. Based on this observation, we propose CMCNet, which aligns image embeddings to fixed textual embeddings via a Center-Margin Contrastive Loss that simultaneously promotes intra-class compactness and inter-class separation. Experimental results show that this embedding alignment strategy is more data-efficient and robust than direct multitask learning, and consistently outperforms InfoNCE, center loss, a strong multitask baseline, and a VQA-style multimodal model, particularly in imbalanced settings. The dataset is freely available at doi: 10.5281/zenodo.19125693 and the source code is available at: https://www.healthinformaticslab.org/supp/.

Materials property prediction remains difficult in low-data settings, where many target properties are supported by only a limited number of labeled samples. Models with the strongest predictive accuracy often depend on crystal structures, which restricts their use in early-stage screening when structural information is limited or unavailable. To address this challenge, we propose DISTAL, a dual-prior framework for structure-agnostic materials property prediction that combines self-supervised compositional pretraining with structure-aware knowledge distillation. DISTAL first learns transferable compositional representations from a large virtual composition space using 145 composition-derived descriptors. It then distills structural knowledge from a pretrained ALIGNN teacher into a composition-conditioned student. This setting allows structural priors to be used during training without requiring structural inputs at inference. By integrating explicit compositional descriptors, pretrained latent features, and distilled structural features within a unified prediction pipeline, DISTAL captures complementary signals that are difficult to recover from any single representation alone. Across 39 benchmark tasks, the best-performing multimodal configuration combines all three signals, and improves over the reference benchmark on 37 tasks. DISTAL achieves the strongest overall performance among all evaluated feature combinations. These results indicate that compositional pretraining and structural distillation provide complementary priors and offer a practical route to robust composition-only prediction in small-data materials informatics. The source code and the pre-trained models are anonymously available at: https://osf.io/eq96d/overview?view_only=451617f42f7849e08750bd1852b48980 and will be released at the official link after acceptance.

Abstract The escalating antimicrobial resistance crisis has intensified the urgent need for alternative antibacterial agents. Phage lysins exhibit potent bactericidal activity with low resistance potential, yet their large-scale discovery from rapidly expanding genomic resources remains limited, as existing computational methods often rely on sequence homology and few reproducible, independently validated, and practically deployable prediction frameworks are currently available. We present LysinFusion, a reproducible deep learning framework that integrates heterogeneous sequence features via a hybrid CNN–Transformer architecture for accurate lysin identification. Trained on a curated, de-redundant corpus (PHROG + inphared) and validated on 148 experimentally confirmed UniProt proteins, LysinFusion achieves an accuracy of 0.81, an AUC of 0.89, and an MCC of 0.62, outperforming the state-of-the-art DeepMineLys by up to 52.3% across metrics while reducing false positives by 64% (12 vs. 33). Ablation studies confirm that both CNN and Transformer modules are individually essential and that serial processing outperforms parallel alternatives. Model interpretability via occlusion and LIME aligns with established lysin biology, with attention concentrating on the N-terminal catalytic domain and decision boundaries reflecting characteristic charge-composition thresholds documented in Gram-negative lysin C-terminal regions. The framework, including all source code, curated datasets, and an accompanying web server, is freely available at https://github.com/sinuo560/LysinFusion .

Sequence-based analysis and prediction form a cornerstone of bioinformatics investigations of the sequence-structure-function paradigm across DNA, RNA, and proteins. The exponential growth in sequence data necessitates sequence encoding methods and advanced predictive models. This study introduces a comprehensive machine learning and deep learning platform SeqHIVE, which aims to streamline the development of prediction pipelines for nucleic acid and protein sequences. SeqHIVE is implemented as a graphical user interface (GUI)–driven platform, enabling intuitive construction, evaluation, and visualization of sequence-based predictive models. SeqHIVE integrates a broad spectrum of 19 biological sequence encoding algorithms, 5 feature selection algorithms, and 26 classifiers for the biological sequence prediction tasks. It automates the processes of sequence-based feature extraction, model construction, predictive performance assessment, statistical analysis, and data visualization. SeqHIVE is engineered to serve both expert bioinformatics researchers with extensive customizable options, and biologists through a user-friendly interface and an intuitive design process. The utility of SeqHIVE is exemplified through a case study on iLearnPlus DNA locus prediction task. The tool SeqHIVE, the documentation, and its source code are freely available at: https://healthinformaticslab.org/supp/. and is also hosted on the GitHub repository: https://github.com/EienKune/SeqHIVE.

Multimodal survival methods that integrate histology whole-slide images and transcriptomic profiles hold significant promise for understanding patient prognostication and guiding personalized treatment strategies. However, existing approaches primarily focus on improving predictive performance through multimodal information fusion, often neglecting the reliability estimation of the prediction results and the inherent alignment noise across modalities. Thus, we propose ReCaSP, a novel and reliable cancer survival prediction framework that effectively integrates histology and transcriptomics data via multimodal alignment and fusion, providing the auxiliary confidence levels for survival predictions through a confidence-aware risk modeling mechanism. Specifically, our approach incorporates a fine-grained risk classifier that models risk labels jointly over both multiple time intervals and censorship status, utilizing evidential deep learning to yield fine-grained risk predictions accompanied by confidence scores. Additionally, to mitigate the inherent noise in multimodal data alignment, we introduce a cross-attention alignment module that effectively aligns histology data with transcriptomics data prior to multimodal fusion, thereby facilitating cross-modal interaction learning. Extensive experiments on five datasets demonstrate that ReCaSP significantly outperforms state-of-the-art methods, achieving a 4.58% improvement in the overall C-Index.

The competing endogenous RNA (ceRNA) regulatory network (CENA) plays a critical role in elucidating the molecular mechanisms of diseases. However, existing computational methods primarily focus on modeling local topological structures of biological networks, struggling to capture high-order regulatory relationships and global topological structures, thus limiting a deeper understanding of complex regulatory interactions. To address this, we propose HpMiX, a Graph Topology-Constrained Mixup (GTCM) and hypergraph residual enhancement learning framework for the discovery of disease-related ceRNA biomarkers. This framework first constructs a CENA network encompassing multi-molecule associations, including miRNA, lncRNA, circRNA, and mRNA, and models higher-order regulatory relationships using K-hop hyperedges. Biologically meaningful initial features are then extracted from CENA via a multi-structure hypergraph weighted random walk method (MHWRW), integrating prior biological knowledge and regulatory information. Subsequently, graph topology-constrained Mixup and multi-head attention, combined with a residual hypergraph neural network, are employed to generate robust node embeddings with both local and global context, enabling the identification of potential disease-ceRNA biomarkers. Prediction results across multiple disease biomarkers demonstrate that HpMiX significantly outperforms state-of-the-art methods, validating its effectiveness in biological regulatory network representation learning. Case studies further confirm that the framework can effectively identify differentially expressed ceRNAs in diseases, highlighting its potential as a tool for pre-screening high-probability disease biomarkers.

In the evolving landscape of drug discovery, the machine learning- and deep learning-based approaches have seen significant growth. Nevertheless, the effectiveness of these algorithms is often impacted by the availability and quality of data. While chemical structure recognition from scientific publications is commonly employed to extract molecular data, the direct recognition of chemical names in the standard IUPAC format from printed or scanned documents remains an underexplored area. We introduce a deep learning framework ChemDT to accurately detect and translate the standard IUPAC names from such sources, subsequently converting them into molecular structures. We also leverage ChemDT to construct the ChemDTD database, encompassing over $\mathbf{3 6, 0 0 0}$ unique small molecules in the implementation of an intuitive graphical user interface. We anticipate that ChemDT holds the potential to assist researchers in the development of comprehensive and high-quality small-molecule database. The software and the ChemDTD database are available at https://github.com/jonechoulover/ChemDT.git.

N7-methylguanosine (m7G) is a common RNA modification linked to multiple diseases. Accurate detection of m7G sites is vital for elucidating its biological roles, but conventional methods are often laborious and costly. AI-based approaches offer alternatives, yet most rely on balanced datasets, ignoring real-world imbalances where negative samples vastly outnumber positives. This discrepancy may lead to overestimated model performance. To address this, we reconstructed independent test sets with low positive-to-negative ratios and benchmarked various models. We propose OCm7G, an ensemble of one-class classifiers with hierarchical thresholding. OCm7G achieves performance comparable to state-of-the-art methods on balanced sets and surpasses them in highly imbalanced scenarios, despite using only 52.5 % of the training data. Moreover, OCm7G offers interpretable predictions, aiding researchers in understanding model decisions. The source code and datasets are publicly available at: https://github.com/lidaosheng/OCm7G.

Recent advances in protein language models have catalyzed significant progress in peptide sequence representation. Despite extensive exploration in this field, pre-trained models tailored for peptide-specific needs remain largely unaddressed due to the difficulty in capturing the complex and sometimes unstable structures of peptides. This study introduces a novel multi-view contrastive learning framework PepHarmony for the sequence-based peptide representation task. PepHarmony innovatively combines sequence- and structure-level information into a sequence-level encoding module through contrastive learning. We carefully select datasets from the Protein Data Bank and AlphaFold DB to encompass a broad spectrum of peptide sequences and structures. The experimental data highlights PepHarmony's exceptional capability in capturing the intricate relationship between peptide sequences and structures compared with the baseline and fine-tuned models. The robustness of our model is confirmed through extensive ablation studies, which emphasize the crucial roles of contrastive loss and strategic data sorting in enhancing predictive performance. The training strategies and the pre-trained PepHarmony model serve as helpful contributions to peptide representations, and offer valuable insights for future applications in peptide drug discovery and peptide engineering.

Introduction In drug discovery, the tacit domain knowledge of experts plays a critical role in guiding molecular design and decision-making.However, existing molecular pre-trained models rarely incorporate such expert knowledge, leading to suboptimal molecular design decisions. Objectives This study proposes MolProphecy, a proxy-human-in-the-loop (proxy-HITL) multi-modal framework that integrates chemists’ domain knowledge with structural molecular information to improve predictive accuracy and interpretability in drug discovery. Methods MolProphecy simulates chemist reasoning using ChatGPT to generate expert-level chemist insights for target molecules.This knowledge is encoded by a large language model (LLM) and fused with graph-based molecular features via a gated multi-head cross-attention module.The framework jointly reasons over human-derived and structural information. Results MolProphecy consistently outperforms baseline models across nine MoleculeNet benchmarks.For illustration, on FreeSolv it achieves an RMSE of 0.796, a 9.1% reduction over the best baseline.On BACE, SIDER, and ClinTox, MolProphecy improves AUROC by 5.39%, 1.43%, and 1.06%, respectively.In addition, MolProphecy demonstrates strong generalization on an independent solubility dataset, demonstrating robustness and applicability of our multi-modal fusion framework. Conclusion MolProphecy provides a generalizable framework for molecular property prediction by integrating simulated chemist expertise with structural data. Its design allows simulated input to be replaced with real chemist knowledge without retraining, establishing a pathway toward collaborative and interpretable drug discovery.

In recent years, the scientific community has become increasingly interested on peptides with non-canonical amino acids due to their superior stability and resistance to proteolytic degradation. These peptides present promising modifications to biological, pharmacological, and physiochemical attributes in both endogenous and engineered peptides. Notwithstanding their considerable advantages, the scientific community exhibits a conspicuous absence of an effective pre-trained model adept at distilling feature representations from such complex peptide sequences. We herein propose PepLand, a novel pre-training architecture for representation and property analysis of peptides spanning both canonical and non-canonical amino acids. In essence, PepLand leverages a comprehensive multi-view heterogeneous graph neural network tailored to unveil the subtle structural representations of peptides. Empirical validations underscore PepLand's effectiveness across an array of peptide property predictions, encompassing protein-protein interactions, permeability, solubility, and synthesizability. The rigorous evaluation confirms PepLand's unparalleled capability in capturing salient synthetic peptide features, thereby laying a robust foundation for transformative advances in peptide-centric research domains. We have made all the source code utilized in this study publicly accessible via GitHub at https://github.com/zhangruochi/pepland

Predicting stock returns remains a central challenge in quantitative finance, transitioning from traditional statistical methods to contemporary deep learning techniques. However, many current models struggle with effectively capturing spatio-temporal dynamics and integrating multiple relational data sources. This study proposes GrifFinNet, a Graph-Relation Integrated Transformer for Financial Predictions, which combines multi-relational graph modeling with Transformer-based temporal encoding. GrifFinNet constructs inter-stock relation graphs based on industry sectors and institutional ownership, and incorporates an adaptive gating mechanism to dynamically integrate relational data in response to changing market conditions. This approach enables the model to jointly capture spatial dependencies and temporal patterns, offering a comprehensive representation of market dynamics. Extensive experiments on two Chinese A-share indices show that GrifFinNet consistently outperforms several baseline models and provides valuable, interpretable insights into financial market behavior. The code and data are available at: https://www.healthinformaticslab.org/supp/.

In today's fast-paced digital communication, the surge in network traffic data and frequency demands robust and precise network intrusion solutions. Conventional machine learning methods struggle to grapple with complex patterns within the vast network intrusion datasets, which suffer from data scarcity and class imbalance. As a result, we have integrated machine learning and deep learning techniques within the network intrusion detection system to bridge this gap. This study has developed TrailGate, a novel framework that combines machine learning and deep learning techniques. By integrating Transformer and Bidirectional Gated Recurrent Unit (BiGRU) architectures with advanced feature selection strategies and supplemented by data augmentation techniques, TrailGate can identifies common attack types and excels at detecting and mitigating emerging threats. This algorithmic fusion excels at detecting common and well-understood attack types and has the unique ability to swiftly identify and neutralize emerging threats that stem from existing paradigms.