Accurate prediction of protein-protein interaction sites is crucial to understanding biological processes, elucidating disease mechanisms, and accelerating drug discovery. Although graph neural network methods have shown potential in this field, but existing methods are limited by the static integrate multi-group features and insufficient perception of hierarchical 3D spatial geometric information, leading to insufficient predictive ability of orphan sites. To address these issues, this paper proposes a Dperception mechanism within a Hierarchical E(n)-equivariant Graph architecture (DHEG). DHEG introduces a dynamic feature importance perception mechanism that adaptively perceives the contextual inter-dependencies of features and assigns weights to feature groups based on their relevance to the interaction relationship. And a hierarchical gated architecture based on E(n)-equivariant graph neural networks that effectively captures protein 3D spatial structures while mitigating over-smoothing problems. The results show that DHEG achieves improvements in 11 of 13 key metrics, with an enhancement 8% in Matthews correlation coefficient, indicating that DHEG not only predicts more interaction sites but also does so with greater reliability. Furthermore, case studies and visualization analyzes show that DHEG aligns better with the biological mechanism and has excellent predictive capabilities for both orphan sites and continuous regions, demonstrating interpretability, and application potential.

Peng Zhou, Pengsen Ma, Jianmin Wang, Xibao Cai, Haitao Huang, Wei Liu, Longyue Wang, Lai Hou Tim, Xiangxiang Zeng. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Molecular Relational Learning (MRL) aims to understand interactions between molecular pairs, playing a critical role in advancing biochemical research. Recently, Large Language Models (LLMs), with their extensive knowledge bases and advanced reasoning capabilities, have emerged as powerful tools for MRL. However, existing LLMs, which primarily rely on SMILES strings and molecular graphs, face two major challenges. They struggle to capture molecular stereochemistry and dynamics, as molecules possess multiple 3D conformations with varying reactivity and dynamic transformation relationships that are essential for accurately predicting molecular interactions but cannot be effectively represented by 1D SMILES or 2D molecular graphs. Additionally, these models do not consider the autoregressive nature of LLMs, overlooking the impact of input order on model performance. To address these issues, we propose DO-CoLM: a Dynamic relationship capture and self-adaptive Ordering 3D molecular Conformation LM for MRL. By introducing modules to dynamically model intra-molecular and inter-molecular conformational relationships and adaptively adjust the molecular modality input order, DO-CoLM achieves superior performance, as demonstrated by experimental results on 12 cross-domain datasets.

Accurate prediction of ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties is crucial in drug development, as these properties directly impact a drug's efficacy and safety. However, existing multi-task learning models often face challenges related to noise interference and task conflicts when dealing with complex molecular structures. To address these issues, we propose a novel multi-task Graph Neural Network (GNN) model, \textbf{MTGIB-UNet}. The model begins by encoding molecular graphs to capture intricate molecular structure information. Subsequently, based on the Graph Information Bottleneck (GIB) principle, the model compresses the information flow by extracting subgraphs, retaining task-relevant features while removing noise for each task. These embeddings are then fused through a gated network that dynamically adjusts the contribution weights of auxiliary tasks to the primary task. Specifically, an uncertainty weighting (UW) strategy is applied, with additional emphasis placed on the primary task, allowing dynamic adjustment of task weights while strengthening the influence of the primary task on model training. Experiments on standard ADMET datasets demonstrate that our model outperforms existing methods. Additionally, the model shows good interpretability by identifying key molecular substructures related to specific ADMET endpoints.

MOTIVATION: Accurate prediction of protein-ligand binding (PLB) relationships plays a crucial role in drug discovery, which helps identify drugs that modulate the activity of specific targets. Traditional biological assays for measuring PLB relationships are time consuming and costly. In addition, models for predicting PLB relationships have been developed and widely used in drug discovery tasks. However, learning more accurate PLB representations is essential to meet the stringent standards required for drug discovery. RESULTS: We propose an image-based PLB representation learning framework, called ImagePLB, which equips ligand representation learner (LRL) and protein representation learner (PRL) to accept 3D multi-view ligand images and protein graphs as input, respectively, and learns rich interaction information between ligand and protein through a binding representation learner (BRL). Considering the scarcity of protein-ligand pairs, we further propose a multi-level next trajectory prediction (MLNTP) task to pre-train ImagePLB on the 4D flexible dynamics trajectory of 16 972 complexes, including ligand level, protein level, and complex level, to learn information related to trajectories. Besides, by introducing trajectory regularization (TR), we effectively alleviate the problem of high (even almost identical) feature similarity caused by adjacent trajectories. Compared with the current state-of-the-art methods, ImagePLB has achieved competitive improvements on PLB-related prediction tasks, including protein-ligand affinity and efficacy prediction tasks. This study opens the door to the image-based PLB learning paradigm. AVAILABILITY AND IMPLEMENTATION: All data and implementation details of code can be obtained from https://github.com/HongxinXiang/ImagePLB.

Accurate prediction of the drug binding between proteins and ligands can significantly advance the development of structure-based drug design. Recent advances have shown great potential in applying equivariant graph neural network (EGNN) -based methods to learn representations of protein-ligand (PL) complexes. However, most of them typically focus on atom-level graph representations and omit the residue-level information in PL complexes, which are considered essential for understanding the binding mechanism. In this article, we develop a SO(3)-equivariant hierarchical graph neural network (EHGNN) that effectively captures the intrinsic hierarchy of biomolecular structures to enhance the predictive performance of PL interactions. Based on the SO(3)-EHGNN, we further propose a molecular dynamics-powered and energy-guided deep learning framework, called Dynamics-PLI, to capture the spatial structures and energetic information inside molecular dynamic (MD) trajectories. Extensive experimental results show significant improvements over current state-of-the-art methods, with a decrease of 4.03% in RMSE for the binding affinity problem and an average increase of 3.95% in AUROC and AUPRC for the ligand efficacy problem, demonstrating the superiority of Dynamics-PLI for PL interaction prediction. Our findings indicate that the SO(3)-EHGNN exhibits enhanced performance without the necessity of pre-training, emphasizing the inherent analytical strength of SO(3)-EHGNN.

Protein-protein interaction (PPI) is a fundamental means of function and signaling in biological systems. The significant increase in demand and cost associated with experimental PPI research requires computational tools to automatically predict and understand PPI. However, existing methods either heavily rely on protein sequences for PPI prediction, or focus on protein structure based on the belief that structure is the key determinant of interactions. But in fact, both have a significant impact on the function of proteins. Therefore, we propose an integrated framework TRGH-PPI based on Transformer and GNN, including a protein feature extraction module and a PPI prediction module, which can simultaneously model both types of protein information and predict PPI. In the protein feature extraction module, we repeatedly use Transformer and GCN to iteratively update the sequence representation and structural features of proteins. The combination of Transformer and GCN enables them to leverage their respective advantages, promote model innovation, and improve the efficiency of graph data processing. In the PPI prediction module, we propose dpdGAT, which uses dot product operations more suitable for PPI prediction and has a dynamic attention mechanism. Numerous experiments have shown that TRGH-PPI is superior to current advanced methods and has shown high accuracy and generalization ability in predicting PPI. By integrating the synergistic modeling of sequence and structural features, the model has achieved an average improvement of 2% -5% in key indicators compared to the SOTA on three datasets.

Disease is one of the primary factors affecting life activities, with complex etiologies often influenced by gene expression and mutation. Currently, wet lab experiments have analyzed the mechanisms of mutations, but these are usually limited by the costs of wet experiments and constraints in sample types and scales. Therefore, this paper constructs a real-world mutation-induced disease dataset and proposes Capsule and Graph topology networks with Multi-head attention (CGM) to predict the mutation-disease associations. CGM can accurately predict protein mutation-disease associations, and to further elucidate the pathogenicity of protein mutations, we also verified that protein mutations lead to protein structural alterations by the model, which suggests that mutation-induced conformational changes may be an important pathogenic factor. Limited by the size of the mutated protein dataset, we also performed experiments on benchmark and imbalanced datasets, where CGM mined 22 unknown protein interaction pairs from the benchmark dataset, better illustrating the potential of CGM in predicting mutation-disease associations. In summary, this paper curates a real dataset. It proposes that CGM predicts protein mutations and disease associations, providing a novel tool for further understanding of biomolecular pathways and disease mechanisms.

Biomedical knowledge graph reasoning is capable of discovering hidden new knowledge based on existing biomedical data, providing ideas and references for new drug discovery, disease research, and so on. The entire graph topology structure formed by triplets and the attribute descriptions for each entity are crucial information for discovering new knowledge. Some add a variety of additional information to aid reasoning, namely multimodal reasoning. However, current multimodal reasoning techniques often rely solely on vector space distance inferences based on triplets themselves, making it difficult to capture more complex relationships and dependencies between facts. This work integrated triplet entity relations, graph topology structures, and attribute descriptions for each entity to incorporate richer information, and utilized logical rules as external knowledge for relational reasoning in biomedical knowledge graphs. We have evaluated our approach on PharmKG.

Peptide drugs are becoming star drug agents with high efficiency and selectivity which open up new therapeutic avenues for various diseases. However, the sensitivity to hydrolase and the relatively short half-life have severely hindered their development. In this study, a new generation artificial intelligence-based system for accurate prediction of peptide half-life was proposed, which realized the half-life prediction of both natural and modified peptides and successfully bridged the evaluation possibility between two important species (human, mouse) and two organs (blood, intestine). To achieve this, enzymatic cleavage descriptors were integrated with traditional peptide descriptors to construct a better representation. Then, robust models with accurate performance were established by comparing traditional machine learning and transfer learning, systematically. Results indicated that enzymatic cleavage features could certainly enhance model performance. The deep learning model integrating transfer learning significantly improved predictive accuracy, achieving remarkable R2 values: 0.84 for natural peptides and 0.90 for modified peptides in human blood, 0.984 for natural peptides and 0.93 for modified peptides in mouse blood, and 0.94 for modified peptides in mouse intestine on the test set, respectively. These models not only successfully composed the above-mentioned system but also improved by approximately 15% in terms of correlation compared to related works. This study is expected to provide powerful solutions for peptide half-life evaluation and boost peptide drug development.

Finding target molecules with specific chemical properties plays a decisive role in drug development. We proposed GEOM-CVAE, a constrained variational autoencoder based on geometric representation for molecular generation with specific properties, which is protein-context-dependent. In terms of machine learning, it includes continuous feature embedding encoder and molecular generation decoder. Our key contribution is to propose an efficient geometric embedding method, including the spatial structure representations of drug molecule (converting the 3-D coordinates into image) and the geometric graph representations of protein target (modeling the protein surface as a mesh). The 3-D geometric information is vital to successful molecular generation, which is different from previous molecular generative methods based on 1-D or 2-D. Our model framework generates specific molecules in two phases, by first generating special image with molecular 3-D information to learn latent representations and generating molecules with constrained condition based on geometric graph convolution for specific protein and then inputting the generated structural molecules into a parser network for obtaining Simplified Molecular Input Line Entry System (SMILES) strings. Our model achieves competitive performance that implies its potential effectiveness to enable the exploration of the vast chemical space for drug discovery.

We construct a protein-protein interaction (PPI) targeted drug-likeness dataset and propose a deep molecular generative framework to generate novel drug-likeness molecules from the features of the seed compounds. This framework gains inspiration from published molecular generative models, uses the key features associated with PPI inhibitors as input and develops deep molecular generative models for de novo molecular design of PPI inhibitors. For the first time, quantitative estimation index for compounds targeting PPI was applied to the evaluation of the molecular generation model for de novo design of PPI-targeted compounds. Our results estimated that the generated molecules had better PPI-targeted drug-likeness and drug-likeness. Additionally, our model also exhibits comparable performance to other several state-of-the-art molecule generation models. The generated molecules share chemical space with iPPI-DB inhibitors as demonstrated by chemical space analysis. The peptide characterization-oriented design of PPI inhibitors and the ligand-based design of PPI inhibitors are explored. Finally, we recommend that this framework will be an important step forward for the de novo design of PPI-targeted therapeutics.

Multi-drug combinations for the treatment of complex diseases are gradually becoming an important treatment, and this type of treatment can take advantage of the synergistic effects among drugs. However, drug-drug interactions (DDIs) are not just all beneficial. Accurate and rapid identifications of the DDIs are essential to enhance the effectiveness of combination therapy and avoid unintended side effects. Traditional DDIs prediction methods use only drug sequence information or drug graph information, which ignores information about the position of atoms and edges in the spatial structure. In this paper, we propose Molormer, a method based on a lightweight attention mechanism for DDIs prediction. Molormer takes the two-dimension (2D) structures of drugs as input and encodes the molecular graph with spatial information. Besides, Molormer uses lightweight-based attention mechanism and self-attention distilling to process spatially the encoded molecular graph, which not only retains the multi-headed attention mechanism but also reduces the computational and storage costs. Finally, we use the Siamese network architecture to serve as the architecture of Molormer, which can make full use of the limited data to train the model for better performance and also limit the differences to some extent between networks dealing with drug features. Experiments show that our proposed method outperforms state-of-the-art methods in Accuracy, Precision, Recall and F1 on multi-label DDIs dataset. In the case study section, we used Molormer to make predictions of new interactions for the drugs Aliskiren, Selexipag and Vorapaxar and validated parts of the predictions. Code and models are available at https://github.com/IsXudongZhang/Molormer.

MOTIVATION: Geometry-based properties and characteristics of drug molecules play an important role in drug development for virtual screening in computational chemistry. The 3D characteristics of molecules largely determine the properties of the drug and the binding characteristics of the target. However, most of the previous studies focused on 1D or 2D molecular descriptors while ignoring the 3D topological structure, thereby degrading the performance of molecule-related prediction. Because it is very time-consuming to use dynamics to simulate molecular 3D conformer, we aim to use machine learning to represent 3D molecules by using the generated 3D molecular coordinates from the 2D structure. RESULTS: We proposed Drug3D-Net, a novel deep neural network architecture based on the spatial geometric structure of molecules for predicting molecular properties. It is grid-based 3D convolutional neural network with spatial-temporal gated attention module, which can extract the geometric features for molecular prediction tasks in the process of convolution. The effectiveness of Drug3D-Net is verified on the public molecular datasets. Compared with other deep learning methods, Drug3D-Net shows superior performance in predicting molecular properties and biochemical activities. AVAILABILITY AND IMPLEMENTATION: https://github.com/anny0316/Drug3D-Net. SUPPLEMENTARY DATA: Supplementary data are available online at https://academic.oup.com/bib.

MOTIVATION: Adverse drug-drug interactions (DDIs) are crucial for drug research and mainly cause morbidity and mortality. Thus, the identification of potential DDIs is essential for doctors, patients and the society. Existing traditional machine learning models rely heavily on handcraft features and lack generalization. Recently, the deep learning approaches that can automatically learn drug features from the molecular graph or drug-related network have improved the ability of computational models to predict unknown DDIs. However, previous works utilized large labeled data and merely considered the structure or sequence information of drugs without considering the relations or topological information between drug and other biomedical objects (e.g. gene, disease and pathway), or considered knowledge graph (KG) without considering the information from the drug molecular structure. RESULTS: Accordingly, to effectively explore the joint effect of drug molecular structure and semantic information of drugs in knowledge graph for DDI prediction, we propose a multi-scale feature fusion deep learning model named MUFFIN. MUFFIN can jointly learn the drug representation based on both the drug-self structure information and the KG with rich bio-medical information. In MUFFIN, we designed a bi-level cross strategy that includes cross- and scalar-level components to fuse multi-modal features well. MUFFIN can alleviate the restriction of limited labeled data on deep learning models by crossing the features learned from large-scale KG and drug molecular graph. We evaluated our approach on three datasets and three different tasks including binary-class, multi-class and multi-label DDI prediction tasks. The results showed that MUFFIN outperformed other state-of-the-art baselines. AVAILABILITY AND IMPLEMENTATION: The source code and data are available at https://github.com/xzenglab/MUFFIN.