The molecular volume, surface area, and polar molecular surface area are important descriptors for characterizing and predicting the molecular properties of lead compounds. Existing computational tools for calculating the above parameters often have complex workflows and are not well-suited for high-throughput conditions. CalVSP is an open-source software for computing molecular volume, molecular surface area, and polar surface area. The software implements a grid-based algorithm that dynamically optimizes grid spacing via quantum chemical reference data to ensure precise parameter calculations. CalVSP was tested on 9489 3D molecular structures, and the results revealed a mean absolute percentage error of 1.25% (95% CI: 1.23–1.27%) for the molecular volume and 1.33% (95% CI: 1.31–1.35%) for the molecular surface area compared with the quantum chemical data. For the molecular polar surface area calculations, the mean absolute percentage error was 4.59% (95% CI: 4.16–5.04%) across the 388 tested molecular structures. The CalVSP written in the C programming language offers a lightweight and easy tool. It can be integrated with other molecular property prediction tools to increase computational accuracy and for large-scale molecular calculations.
Accurate and interpretable modeling of crystalline materials is essential for understanding the structure-property relationships in materials critical in accelerating materials discovery. While recent graph neural networks (GNNs) have achieved high predictive accuracy, they often struggle to provide physical interpretability and fail to explicitly model the hierarchical and symmetrical nature of crystals. In this work, we introduce Capsule Graph Networks with E(3)-Equivariance (CGN-e3), a novel deep learning framework that integrates equivariant message passing with capsule networks to capture both geometric symmetries and hierarchical motif structures. CGN-e3 leverages E(3)-equivariant message passing to learn physically consistent features and organize them into capsule representations that can disentangle local motifs, such as polyhedral environments, and connects them to global properties. We validate the effectiveness of our framework on bandgap and formation energy prediction, as well as material classification using Materials Project and Matbench datasets. Our model achieves a MAE of 0.054 eV/atom and 0.379 eV on formation energy and bandgap prediction, respectively, outperforming CGCNN and matching the performance of MEGNet on the same dataset, while also providing insightful interpretations of the learned capsule representations.Scientific contribution: We present the first integration of E(3)-equivariant graph neural networks with capsule networks for modeling crystalline materials. This unified architecture captures both the fundamental physical symmetries of 3D space; rotation, translation, reflection and the intrinsic hierarchical part-whole relationships e.g., atoms to polyhedra to extended motifs found in crystal structures. The framework provides an unsupervised pathway for interpretable motif discovery. The dynamic routing-by-agreement mechanism identifies and aggregates structurally significant local environments such as the octahedra into higher-order graph-level capsules. This process yields human-intelligible insights by explicitly quantifying the contribution of specific structural motifs to target material properties, moving beyond "black-box" predictions. We validate our framework on key property prediction tasks and provide capsule-level interpretation of the results.
Hundreds of computational methods for predicting ligand binding pockets exist, but the problem of finding druggable pockets throughout the human proteome persists. Different strategies for pocket-finding excel in different use cases. Ensemble models that leverage multiple different pocket-finding strategies can best capture diverse pockets at scale. Despite this, no publicly available human-proteome-wide datasets of pocket predictions from multiple pocket-finding methods exist. We present the Human Omnibus of Targetable Pockets (HOTPocket), a dataset of over 2.4 million predicted pockets over the entire human proteome that utilizes both experimentally-determined and computationally-predicted protein structures. We assembled this dataset by running seven diverse, established pocket-finding methods over all PDB and AlphaFold2 structures of the canonical human proteome. We created a novel pocket scoring method, hotpocketNN, which we used to filter candidate pockets and assemble the final proteome-wide dataset. Our hotpocketNN method is able to recover known ligand binding pockets, including those which are dissimilar from any pocket seen in its training set. The hotpocketNN method outperforms all constituent methods, including P2Rank and Fpocket, when assessing the precision with DCA criterion on the Astex Diverse Set and PoseBusters dataset. Additionally, hotpocketNN was able to identify recently-discovered druggable pockets on KRAS and the mu opioid receptor. We make both the HOTPocket dataset and the hotpocketNN method freely available.
Caco-2= permeability serves as a critical in vitro indicator for predicting the oral absorption of drug candidates= during early-stage drug discovery. To enhance the accuracy and= efficiency of computational predictions, we systematically investigated the impact of eight molecular feature= representation types including 2D/3D descriptors, structural fingerprints, and deep learning-based embeddings combined with automated machine learning techniques to predict Caco-2 permeability. We evaluated model performance across various molecular representations using two datasets differing in scale and chemical diversity, namely the TDC benchmark and curated OCHEM data. Among the tested fingerprints and descriptors, PaDEL, Mordred, and RDKit emerged as particularly effective for predicting Caco-2 permeability. Notably, our model CaliciBoost, identified through training optimization, achieved the lowest MAE and secured the top position on the TDC Caco-2 Leaderboard. Furthermore, for both Padel and Mordred, using TDC data, incorporating 3D descriptors seem lead to improvements over using 2D features alone, as supported by feature importance analyses. These findings highlight the effectiveness of automated machine learning approaches in ADMET modeling and offer practical guidance for feature selection in data-limited prediction tasks.
This work provides a systematic benchmarking of eight molecular feature representation types in conjunction with AutoML for Caco-2 permeability prediction. It highlights the critical role of 3D descriptors in enhancing predictive accuracy and establishes a PaDEL-based AutoML model that achieves top-ranked performance on a public leaderboard. The study also emphasizes the value of interpretable feature selection (via SHAP and permutation importance), offering insights into feature contributions and generalizable modeling strategies for cheminformatics applications.
Understanding molecular interactions beyond single-molecule properties is critical for studying real-world chemical systems. Quantum chemical calculations of molecule–molecule interactions are computationally demanding, making large, publicly available datasets scarce. Here, we present an efficient framework for generating initial configurations of molecular interaction systems and construct a molecular interaction dataset, containing 49,620 individual molecules and 247,741 molecular pairs spanning chromophore–solvent, solute–solvent, and drug–drug interactions, each associated with experimentally characterized equilibrium structures. Our dataset can be used for theoretical studies and machine learning applications in chemical sciences, particularly for modeling intermolecular interactions and structure-based prediction of experimental properties. In future work, we plan to expand the dataset to include non-equilibrium structures and atomic forces, thereby broadening its applicability to reaction modeling and force field development.
Molecular generation is a critical method in drug design, but its practical application is often limited by the difficulty of synthesizing the generated molecules. To address this challenge, we present RetroScore, a synthetic accessibility evaluation framework guided by multistep retrosynthesis. Our methodology integrates the semi-template model Graph2Edits with the multistep retrosynthesis planning algorithm Retro*, forming the Graph2Edits-Retro*d system. By incorporating the green chemistry metric of graph edit distance into the reaction cost function and a multistage screening protocol, this system identifies optimal routes while balancing reliability, synthetic efficiency, and economic feasibility. Benchmark evaluations demonstrate a 97.37% planning success rate with balanced optimization across route length, confidence score, and graph edit distance. In the molecular generation task, the RetroScore outperforms six of the seven synthetic accessibility metrics, yielding molecules with enhanced synthetic accessibility profiles across heterogeneous evaluation frameworks. To facilitate practical implementation, we developed an open-access web platform for automated retrosynthesis route prediction and RetroScore calculation, providing researchers with rapid synthetic accessibility assessments. The RetroScore web server is publicly accessible at http://aidd.bioai-global.com/RetroScore/, and the source code is available at https://github.com/Snowgao320/RetroScore.
Peptide precursors, as the source molecules of bioactive peptides, play essential roles in neuroregulation, immune defense, and drug development. Their accurate identification is crucial for elucidating mechanisms of life regulation and developing novel therapeutics. However, the complexity and diversity of peptide precursor sequences pose significant challenges to prediction tasks. Existing methods predominantly rely on sequence features or structural features, hindering the full exploitation of complementary information between modalities and consequently limiting prediction performance. We introduce ProjFusNet, a deep learning framework that integrates evolutionary-scale protein sequence representations from ESM-2 with structural features via a projected multimodal fusion strategy. A bidirectional LSTM is further employed to model the complex interactions between sequence and structure. In rigorous five-fold cross-validation, ProjFusNet demonstrates improved performance across key metrics, including ACC, SN, AUC, SP, and MCC, compared to single-feature models.
The knowledge panels in PubChem allow users to quickly identify and summarize important relationships between chemicals, genes, proteins, and diseases by analyzing the co-occurrences of those entities in a collection of text documents. In the present study, the analysis and summarization techniques used to develop the literature knowledge panels in PubChem were extended to patent documents from the Google Patent Research Data (GPRD) set. The annotations of the patent documents in the GPRD set were mapped to NCBI database records to create the patent co-occurrence data. The annotations were not only from the titles and abstracts of patents but also from other parts such as claims and descriptions, greatly improving the coverage of the co-occurrence-based entity relationships in PubChem. Informativeness weights of entities were introduced in the co-occurrence and relevance score computations to account for a significant variation in the number of matched annotations per patent section. This narrows the focus to the co-occurrences that are more relevant to the subject matter of the patent. The resulting co-occurrence data was used to generate the patent knowledge panels, enabling users to identify entities co-mentioned in patents alongside a specific chemical or gene. The patent co-occurrence data can be downloaded interactively or accessed programmatically. Overall, the patent knowledge panels described in this study provide users with quick access to essential biomedical entities associated with a given PubChem record. Users can delve into relevant patent documents related to these entities or download the underlying co-occurrence data for further exploration and analysis.
We present VN-EGNN, a novel approach to binding site identification that significantly advances predictive performance. By integrating virtual nodes into E(n)– and SE(n)-equivariant graph neural networks (EGNNs) and extending the message-passing scheme, we address limitations of traditional GNNs in modeling complex geometric entities such as binding pockets and at the same time get neural representations of binding sites. Our extensive experiments demonstrate that VN-EGNN sets a new state-of-the-art in locating binding site centers on the COACH420, HOLO4K, and PDBbind2020 datasets, showcasing a marked improvement in the DCC/DCA success rates over existing methods. These results underscore the potential of VN-EGNN in drug discovery and protein-ligand interaction studies.
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：