Journal of Chemical Information and Modeling 最新文献

Fullerenes are hollow carbon molecules where each atom is connected to exactly three other atoms, arranged in pentagonal and hexagonal rings. Mathematically, they can be combinatorially modeled as planar, 3-regular graphs with facets composed only of pentagons and hexagons. In this work, we outline a few of the many open questions about fullerenes, beginning with the problem of generating fullerenes randomly. We then introduce an infinite family of fullerenes on which the generalized Stone-Wales operation is inapplicable. Furthermore, we present numerical insights into a graph invariant, called the character of a fullerene, derived from its adjacency and degree matrices. As supported by numerical results, this descriptor may lead to a new method for linear enumeration of all fullerenes.

Bioactive peptides from food sources offer a safe and biocompatible approach to enzyme inhibition, with potential applications in managing metabolic disorders such as hyperuricemia and gout, conditions linked to excessive xanthine oxidase activity. Using a machine learning-based screening approach inspired by the bioactivity of natto, two peptides, ECFK and FECK, were identified from the Bacillus subtilis proteome and validated as xanthine oxidase inhibitors with IC₅₀ values of 37.36 and 71.57 mM, respectively. Further experiments confirmed their safety through cytotoxicity assays, and electronic tongue analysis demonstrated their mild sensory properties, supporting their edibility. Molecular dynamics simulations revealed that these peptides stabilize critical enzyme regions, with ECFK showing a higher dissociation energy barrier (52.08 kcal/mol) than FECK (46.39 kcal/mol), indicating strong, stable interactions. This study highlights food-derived peptides as safe and natural inhibitors of xanthine oxidase, offering promising therapeutic potential for metabolic disorder management.

The high doses of drugs required for biotherapeutics, such as monoclonal antibodies (mAbs), and the small volumes that can be administered to patients by subcutaneous injections pose challenges due to high-concentration formulations. The addition of excipients, such as arginine and glutamate, to high-concentration protein formulations can increase solubility and reduce the tendency of protein particle formation. Molecular dynamics (MD) simulations can provide microscopic insights into the mode of action of excipients in mAb formulations but require large system sizes and long time scales that are currently beyond reach at the fully atomistic level. Computationally efficient coarse-grained models such as the Martini 3 force field can tackle this challenge but require careful parametrization, testing, and validation. This study extends the popular Martini 3 force field toward realistic protein-excipient interactions of arginine and glutamate excipients, using the Fab domains of the therapeutic mAbs trastuzumab and omalizumab as model systems. A novel all-atom to coarse-grained mapping of the amino acid excipients is introduced, which explicitly captures the zwitterionic character of the backbone. The Fab-excipient interactions of arginine and glutamate are characterized concerning molecular contacts with the Fabs at the single-residue level. The Martini 3 simulations are compared with results from all-atom simulations as a reference. Our findings reveal an overestimation of Fab-excipient contacts with the default interaction parameters of Martini 3, suggesting a too strong attraction between protein residues and excipients. Therefore, we reparametrized the protein-excipient interaction parameters in Martini 3 against all-atom simulations. The excipient interactions obtained with the new Martini 3 mapping and Lennard-Jones (LJ) interaction parameters, coined Martini 3-exc, agree closely with the all-atom reference data. This work presents an improved parameter set for mAb-arginine and mAb-glutamate interactions in the Martini 3 coarse-grained force field, a key step toward large-scale coarse-grained MD simulations of high-concentration mAb formulations and the stabilizing effects of excipients.

Expanding enzyme substrate spectra enhances industrial applications and drives sustainable biocatalysis. Despite advances, challenges in modification efficiency and high-throughput screening persist. Here, we developed a virtual screening method called CMDmpnn that combines comparative molecular dynamics (MD) simulations and ProteinMPNN to broaden enzyme substrate spectra without compromising other industrially important properties of enzymes, such as thermostability. Using glycosyltransferase as a model, we first established a dynamic model library of the wild-type enzyme through MD simulations and performed clustering. Subsequently, we utilized ProteinMPNN to generate a comprehensive set of new sequences for the entire library, enabling rapid identification of all possible enzyme variants. Short MD simulations were then conducted on variant-substrate complex models, with results compared to those of the wild-type enzyme. By analyzing catalytically relevant information such as substrate binding modes and key atomic distances, we identified multiple variants capable of catalyzing a broad spectrum of phenolic compounds, all within a timeframe of less than 2 weeks. The CMDmpnn method offers a powerful and efficient tool for rapidly expanding enzyme substrate spectra.

The application of large language models in materials science has opened new avenues for accelerating materials development. Building on this advancement, we propose a novel framework leveraging large language models to optimize experimental procedures for synthesizing quantum dot materials with multiple desired properties. Our framework integrates the synthesis protocol generation model and the property prediction model, both fine-tuned on open-source large language models using parameter-efficient training techniques with in-house synthesis protocol data. Once the synthesis protocol with target properties and a masked reference protocol is generated, it undergoes validation through the property prediction models, followed by assessments of its novelty and human evaluation. Our synthesis experiments demonstrate that among the six synthesis protocols derived from the entire framework, three successfully update the Pareto front, and all six improve at least one property. Through empirical validation, we confirm the effectiveness of our fine-tuned large language model-driven framework for synthesis planning, showcasing strong performance under multitarget optimization.

A complex web of intermolecular interactions defines and regulates biological processes. Understanding this web has been particularly challenging because of the sheer number of actors in biological systems: ∼10⁴ proteins in a typical human cell offer plausible 10⁸ interactions. This number grows rapidly if we consider metabolites, drugs, nutrients, and other biological molecules. The relative strength of interactions also critically affects these biological processes. However, the small and often incomplete data sets (10³–10⁴ protein–ligand interactions) traditionally used for binding affinity predictions limit the ability to capture the full complexity of these interactions. To overcome this challenge, we developed Yuel 2, a novel neural network-based approach that leverages transfer learning to address the limitations of small data sets. Yuel 2 is pretrained on a large-scale data set to learn intricate structural features and then fine-tuned on specialized data sets like PDBbind to enhance the predictive accuracy and robustness. We show that Yuel 2 predicts multiple binding affinity metrics, K_d, K_i, and IC₅₀, between proteins and small molecules, offering a comprehensive representation of molecular interactions crucial for drug design and development.

In the quest for accelerating de novo drug discovery, the development of efficient and accurate scoring functions represents a fundamental challenge. This study introduces iScore, a novel machine learning (ML)-based scoring function designed to predict the binding affinity of protein-ligand complexes with remarkable speed and precision. Uniquely, iScore circumvents the conventional reliance on explicit knowledge of protein-ligand interactions and a full picture of atomic contacts, instead leveraging a set of ligand and binding pocket descriptors to directly evaluate binding affinity. This approach enables skipping the inefficient and slow conformational sampling stage, thereby enabling the rapid screening of ultrahuge molecular libraries, a crucial advancement given the practically infinite dimensions of chemical space. iScore was rigorously trained and validated using the PDBbind 2020 refined set, CASF 2016, CSAR NRC-HiQ Set1/2, DUD-E, and target fishing data sets, employing three distinct ML methodologies: Deep neural network (iScore-DNN), random forest (iScore-RF), and eXtreme gradient boosting (iScore-XGB). A hybrid model, iScore-Hybrid, was subsequently developed to incorporate the strengths of these individual base learners. The hybrid model demonstrated a Pearson correlation coefficient (R) of 0.78 and a root-mean-square error (RMSE) of 1.23 in cross-validation, outperforming the individual base learners and establishing new benchmarks for scoring power (R = 0.814, RMSE = 1.34), ranking power (ρ = 0.705), and screening power (success rate at top 10% = 73.7%). Moreover, iScore-Hybrid demonstrated great performance in the target fishing benchmarking study.

Nitroaromatic compounds (NAs) are widely used in industrial applications but pose significant genotoxic risks, necessitating accurate mutagenicity prediction for chemical safety assessments. This study integrates conceptual density functional theory (CDFT) descriptors with explainable no-code machine learning (ML) models to predict NA mutagenicity based on Ames test results. Following OECD QSAR guidelines, feature selection and model development were performed using decision-tree-based algorithms (Random Tree, JCHAID*, SPAARC) and multilayer perceptrons (MLPs). These models exhibited high predictive accuracy (internal: >80%, κ = 0.21-0.37; external: ∼90%, κ = 0.41-0.62) with strong interpretability. The study also explores the role of metabolic activation and aqueous-phase descriptors, evaluating a novel electronic analog to LogP (LogQP) to assess hydrophobicity-mutagenicity relationships. Results demonstrate that aqueous-phase electronic properties and electrophilicity descriptors outperform vacuum-based methods in mutagenicity prediction. The combination of CDFT descriptors with shallow ML models proves to be a robust, interpretable, and accessible framework for predictive toxicology. This approach enhances chemical risk assessment and bridges computational chemistry with toxicology for regulatory applications.

Drug-drug interactions influence drug efficacy and patient prognosis, providing substantial research value. Some existing methods struggle with the challenges posed by sparse networks or lack the capability to integrate data from multiple sources. In this study, we propose MOLGAECL, a novel approach based on graph autoencoder pretraining and molecular graph contrastive learning. Initially, a large number of unlabeled molecular graphs are pretrained using a graph autoencoder, where graph contrastive learning is applied for more accurate representation of the drugs. Subsequently, a full-parameter fine-tuning is performed on different data sets to adapt the model for drug interaction-related prediction tasks. To assess the effectiveness of MOLGAECL, comparison experiments with state-of-the-art methods, fine-tuning comparison experiments, and parameter sensitivity analysis are conducted. Extensive experimental results demonstrate the superior performance of MOLGAECL. Specifically, MOLGAECL achieves an average increase of 6.13% in accuracy, 6.14% in AUROC, and 8.16% in AUPRC across all data sets.

Viruses are significant human pathogens responsible for pandemic outbreaks and seasonal epidemics. Viral infectious diseases impose a devastating global burden and have a profound impact on public health systems. During viral infections, alternative splicing (AS) plays a crucial role in regulating immune responses, altering the host's cellular environment, expanding viral genetic material, and facilitating viral replication. As research on AS in viral infections expands, it is crucial to consolidate data on virus-related splicing changes to improve our understanding of these viruses and associated diseases. To address this need, we created ASVirus (https://bddg.hznu.edu.cn/asvirus/), a comprehensive database of virus-associated AS events and their regulatory factors. ASVirus uniquely combines high-confidence, experimentally validated splicing data and investigates upstream regulatory mechanisms through a gene-splicing factor interaction network. Its user-friendly web interface offers detailed information into AS events from various viral families and the resulting mis-splicing in host genes, aiding the exploration of novel viral infection mechanisms and the identification of critical therapeutic targets for viral diseases.