Aug 2026· Applied Sciences· 0 citations· 38 references
TL;DR
DHST is proposed, a deep hybrid structure–topology framework that integrates sequence semantics from a pretrained protein language model with local structural information learned by a residual graph convolutional network and introduces site-specific persistent homology to encode multi-scale topological invariants and a topology-guided residue-wise gated fusion module to modulate structure–semantics representations using local topological embeddings.
Abstract
Accurate protein function prediction (PFP) is essential for understanding biological systems. However, structure-based graph neural networks often rely on fixed-distance contact maps, which may inadequately capture continuous, multi-scale spatial topologies, while the long-tail distribution of Gene Ontology (GO) labels may bias prediction toward frequent functions. We propose DHST, a deep hybrid structure–topology framework that integrates sequence semantics from a pretrained protein language model with local structural information learned by a residual graph convolutional network. DHST further introduces site-specific persistent homology to encode multi-scale topological invariants and a topology-guided residue-wise gated fusion module to modulate structure–semantics representations using local topological embeddings. The fused residue features are aggregated through dual-path pooling, and a weighted binary cross-entropy loss is used to mitigate the adverse effects of label imbalance. On the PDB dataset, DHST achieved area under the precision–recall curve (AUPR) scores of 0.779, 0.481, and 0.557 for molecular function (MF), biological process (BP), and cellular component (CC), respectively; on the AF2 dataset, the corresponding scores were 0.729, 0.390, and 0.459. The model also demonstrated robust generalization to low-homology proteins and maintained strong predictive performance across GO terms with different levels of functional specificity. Ablation results supported the contributions of the main components.
By combining protein language model embeddings with topology-adaptive geometric reasoning, DiConSite offers a reusable framework for residue-level protein interaction analysis and achieves consistently strong and often best-performing results, while improving robustness to structural uncertainty and cross-modal variation.
Shou-Zhi Chen, Zhenchao Tang, Linlin You et al.· IEEE Transactions on Pattern...· 1 citation
This work presents HGRL-PPIS, a novel hierarchical graph representation learning approach for predicting protein-protein interaction sites that achieves superior performance over competing methods on multiple benchmark datasets, enabling more reliable detection of protein-protein binding residues.
This work proposes ARF-GNN, an adaptive receptive field graph neural network tailored for protein function prediction, which dynamically models structural context via hierarchical multi-hop neighborhood aggregation and introduces a dual-branch meta-learning framework.
Zhiqiang Hui, Weizhong Lu, Yiyi Xia et al.· Computational biology and ch...· 0 citations
Sirtuins comprise a group of proteins that play critical roles in regulating gene expression, DNA repair, metabolic homeostasis, cellular stress responses, apoptosis, and aging-related pathways. Accurate classification of sirtuin proteins is important for understanding biological functions and drug development. While foundational models such as DeepSIRT achieved robust performance using one-dimensional Convolutional Neural Networks (1D-CNN) paired with traditional features like PSSM and AAC, these approaches primarily rely on static, manually-engineered representations. Such methods often fail to capture the deep contextual semantics hidden in protein sequences or the dynamic spatial topologies essential for functional specificity. Furthermore, these sequence-centric models remain 'structurally blind', as they lack a mechanism to distinguish between high-confidence functional domains and disordered, non-informative regions. To address these limitations, we introduce SirtSAGE (Sirtuin Structural-Aware Graph-Evolutionary framework), a hybrid architecture designed to distill evolutionary embeddings through physical structural constraints. SirtSAGE integrates the contextual semantics of ESM-2 into a Graph Attention Network (GATv2), where pLDDT scores serve as spatial confidence gates to fine-tune the influence of structural interactions. To bridge the gap between high performance and transparency, we employ Kolmogorov-Arnold Networks (KAN) as a symbolic reasoning layer, enabling the extraction of non-linear functional signatures that are often obscured in standard black-box MLPs. Extensive evaluations on both 5-fold cross-validation and an independent test set demonstrate that SirtSAGE outperforms state-of-the-art 1D-CNN baselines in predictive robustness, despite operating under a substantially more stringent unified multi-class setting. Our analysis reveals that SirtSAGE effectively concentrates attention on the structurally stable 'Active Core' of sirtuins while organically learning a continuous latent space whose spatial organization is qualitatively consistent with the established 4-class phylogenetic classification of mammalian sirtuins. Furthermore, successful orthogonal validation on C. elegans variants highlights the framework's mathematical rigour for cross-species homology mapping. SirtSAGE represents a paradigm shift from implicit pattern recognition toward structurally-grounded, interpretable protein annotation.
Dinh-Quy Nguyen, Viet-Thanh Nguyen, Muhammad Hussain et al.· Journal of Computational Bio...· 0 citations
The novel combination of LZ complexity–based negative sample selection, CT feature representation, and GA-optimized CNN–LSTM architecture provides a robust and biologically informed framework for PPI prediction.
Identifying transcription factor binding sites (TFBSs) is fundamental to understanding complex gene regulatory mechanisms and the functions of non-coding regions. Although existing methods have achieved substantial strides, capturing both local structural features and long-range spatial dependencies within DNA sequences remains a major challenge for improving prediction accuracy. In this study, we propose DNCLA, a deep learning model that synergizes multisize convolutional fusion, Bidirectional Long ShortTerm Memory (Bi-LSTM) networks, and a multi-head self-attention
mechanism. At the feature extraction level, DNCLA breaks through the limitations of traditional single-sequence encoding by fusing Nucleotide Chemical Properties (NCP) with Dinucleotide Physicochemical Properties (DPCP). NCP provides a refined characterization of chemical differences between bases based on ring structures, hydrogen bond sites, and functional group properties, while DPCP introduces parameters such as local structural stability and geometric flexibility of the DNA. Subsequently, the model extracts spatial evolution from these high-dimensional features through a multi-size convolutional module; captures long-range spatial dependencies using Bi-LSTM layers; and employs a multi-head self-attention mechanism to achieve adaptive weight distribution of global features, thereby enhancing the perception of key regulatory motifs. Results from training and testing the proposed model on 165 ChIPseq datasets demonstrate that DNCLA possesses robust generalization capabilities and high predictive performance in TFBSs identification. This suggests that the incorporation of physicochemical features better elucidates the essence of interactions between transcription factors and DNA.
Jingjue Wei, Jie Feng· Match-communications in Math...· 0 citations