Journal of Chemical Information and Modeling 最新文献

Kinases are critical regulators in numerous cellular processes, and their dysregulation is linked to various diseases, including cancer. Thus, protein kinases have emerged as major drug targets at present, with approximately a quarter to a third of global drug development efforts focusing on kinases. Additionally, deep learning-based molecular generation methods have shown obvious advantages in exploring large chemical space and improving the efficiency of drug discovery. However, many current molecular generation models face challenges in considering specific targets and generating molecules with desired properties, such as target-related activity. Here, we developed a specialized and enhanced deep learning-based molecular generation framework named KinGen, which is specially designed for the efficient generation of small molecule kinase inhibitors. By integrating reinforcement learning, transfer learning, and a specialized reward module, KinGen leverages a binding affinity prediction model as part of its reward function, which allows it to accurately guide the generation process toward biologically relevant molecules with high target activity. This approach not only ensures that the generated molecules have desirable binding properties but also improves the efficiency of molecular optimization. The results demonstrate that KinGen can generate structurally valid, unique, and diverse molecules. The generated molecules exhibit binding affinities to the target that are comparable to known inhibitors, achieving an average docking score of -9.5 kcal/mol, which highlights the model's ability to design compounds with enhanced activity. These results suggest that KinGen has the potential to serve as an effective tool for accelerating kinase-targeted drug discovery efforts.

Intrinsically disordered proteins (IDPs) have garnered significant attention due to their critical roles in complex human diseases. Molecular dynamics (MD) simulations have emerged as a valuable approach for studying IDPs, whose accuracy heavily depends on the accuracy of force fields. Despite this, the high conformational flexibility of IDPs presents limitations for current force fields in precisely capturing their conformational features. Here, we developed a tool for generating force field parameters, consisting of two main components: the construction and training of a model named DihedralProbNet to predict protein dihedral probability distributions and the DeepReweighting algorithm to optimize force field parameters. This personalized energy adaptation through reweighting learning was termed the PEARL force field. To evaluate its performance, 8 IDPs and 5 folded protein systems were used. The results demonstrate that the PEARL force field more accurately reproduces the conformational ensembles of IDPs than ff19SB and stabilizes the conformations of folded proteins. Therefore, by enabling a more accurate sampling of IDP conformations, PEARL has the potential to advance our understanding of the role of IDPs in biological processes and their involvement in disease mechanisms.

Intrinsically disordered proteins (IDPs) have garnered significant attention due to their critical roles in complex human diseases. Molecular dynamics (MD) simulations have emerged as a valuable approach for studying IDPs, whose accuracy heavily depends on the accuracy of force fields. Despite this, the high conformational flexibility of IDPs presents limitations for current force fields in precisely capturing their conformational features. Here, we developed a tool for generating force field parameters, consisting of two main components: the construction and training of a model named DihedralProbNet to predict protein dihedral probability distributions and the DeepReweighting algorithm to optimize force field parameters. This personalized energy adaptation through reweighting learning was termed the PEARL force field. To evaluate its performance, 8 IDPs and 5 folded protein systems were used. The results demonstrate that the PEARL force field more accurately reproduces the conformational ensembles of IDPs than ff19SB and stabilizes the conformations of folded proteins. Therefore, by enabling a more accurate sampling of IDP conformations, PEARL has the potential to advance our understanding of the role of IDPs in biological processes and their involvement in disease mechanisms.

Alzheimer’s disease (AD) is a complex neurodegenerative disorder with an unclear pathogenesis; the traditional ″single gene-single target-single drug″ strategy is insufficient for effective treatment. This study explores a novel strategy for the multitarget therapy of AD by integrating multiomics data and employing network analysis. Different from conventional single-target methods, TCnet adopts a mechanism-driven strategy, utilizing multiomics data to decompose disease mechanisms, construct potential target combinations, and prioritize the optimal combinations using a scoring function. TCnet not only advances our understanding of disease mechanisms but also facilitates large-scale drug screening. This approach was further employed to screen active compounds from Huang-Lian-Jie-Du-Tang (HLJDT), identifying quercetin as a candidate targeting GSK3β and ADAM17. Subsequent in vitro experiments confirmed the neuroprotective and anti-inflammatory effects of quercetin. Overall, TCnet offers a promising approach for predicting target combinations and provides new insights and directions for drug discovery in AD.

Alzheimer's disease (AD) is a complex neurodegenerative disorder with an unclear pathogenesis; the traditional ″single gene-single target-single drug″ strategy is insufficient for effective treatment. This study explores a novel strategy for the multitarget therapy of AD by integrating multiomics data and employing network analysis. Different from conventional single-target methods, TCnet adopts a mechanism-driven strategy, utilizing multiomics data to decompose disease mechanisms, construct potential target combinations, and prioritize the optimal combinations using a scoring function. TCnet not only advances our understanding of disease mechanisms but also facilitates large-scale drug screening. This approach was further employed to screen active compounds from Huang-Lian-Jie-Du-Tang (HLJDT), identifying quercetin as a candidate targeting GSK3β and ADAM17. Subsequent in vitro experiments confirmed the neuroprotective and anti-inflammatory effects of quercetin. Overall, TCnet offers a promising approach for predicting target combinations and provides new insights and directions for drug discovery in AD.

Kinases are critical regulators in numerous cellular processes, and their dysregulation is linked to various diseases, including cancer. Thus, protein kinases have emerged as major drug targets at present, with approximately a quarter to a third of global drug development efforts focusing on kinases. Additionally, deep learning-based molecular generation methods have shown obvious advantages in exploring large chemical space and improving the efficiency of drug discovery. However, many current molecular generation models face challenges in considering specific targets and generating molecules with desired properties, such as target-related activity. Here, we developed a specialized and enhanced deep learning-based molecular generation framework named KinGen, which is specially designed for the efficient generation of small molecule kinase inhibitors. By integrating reinforcement learning, transfer learning, and a specialized reward module, KinGen leverages a binding affinity prediction model as part of its reward function, which allows it to accurately guide the generation process toward biologically relevant molecules with high target activity. This approach not only ensures that the generated molecules have desirable binding properties but also improves the efficiency of molecular optimization. The results demonstrate that KinGen can generate structurally valid, unique, and diverse molecules. The generated molecules exhibit binding affinities to the target that are comparable to known inhibitors, achieving an average docking score of −9.5 kcal/mol, which highlights the model’s ability to design compounds with enhanced activity. These results suggest that KinGen has the potential to serve as an effective tool for accelerating kinase-targeted drug discovery efforts.

Allosteric compounds offer an alternative mode of inhibition to orthosteric compounds with opportunities for selectivity and noncompetition. Structure-based drug design (SBDD) of allosteric compounds introduces complications compared to their orthosteric counterparts; multiple binding sites of interest are considered, and often allosteric binding is only observed in particular protein conformations. Blind docking methods show potential in virtual screening allosteric ligands, and deep learning methods, such as DiffDock, achieve state-of-the-art performance on protein-ligand complex prediction benchmarks compared to traditional docking methods such as Vina and Lin_F9. To this aim, we explore the utility of a data-driven platform called the minimum distance matrix representation (MDMR) to retrospectively predict recently discovered allosteric inhibitors complexed with Cyclin-Dependent Kinase (CDK) 2. In contrast to other protein complex representations, it uses the minimum residue-residue (or residue-ligand) distance as a feature that prioritizes the formation of interactions. Analysis of this representation highlights the variety of protein conformations and ligand binding modes, and we identify an intermediate protein conformation that other heuristic-based kinase conformation classification methods do not distinguish. Next, we design self- and cross-docking benchmarks to assess whether docking methods can predict both orthosteric and allosteric binding modes and if prospective success is conditional on the selection of the protein receptor conformation, respectively. We find that a combined method, DiffDock followed by Lin_F9 Local Re-Docking (DiffDock + LRD), can predict both orthosteric and allosteric binding modes, and the intermediate conformation must be selected to predict the allosteric pose. In summary, this work highlights the value of a data-driven method to explore protein conformations and ligand binding modes and outlines the challenges of SBDD of allosteric compounds.

In this work, a reliable and robust trust region method for restricted minimizations on hyperspheres is developed. The working equations of this new methodology are presented, together with their validation. The performance and characteristics of this new algorithm are discussed by a constrained minimization on a sphere using a two-dimensional Quapp model surface. The obtained results show that the proposed method for minimizations on hyperspheres guarantees convergence to constrained minima. Its enhanced numerical stability permits tight convergence criteria for constrained minimizations. The application of the new restricted minimizer in the framework of the hierarchical transition state finder and for the calculation of intrinsic reaction coordinates for 38 chemical reactions demonstrates its robustness and efficiency.

Allosteric compounds offer an alternative mode of inhibition to orthosteric compounds with opportunities for selectivity and noncompetition. Structure-based drug design (SBDD) of allosteric compounds introduces complications compared to their orthosteric counterparts; multiple binding sites of interest are considered, and often allosteric binding is only observed in particular protein conformations. Blind docking methods show potential in virtual screening allosteric ligands, and deep learning methods, such as DiffDock, achieve state-of-the-art performance on protein–ligand complex prediction benchmarks compared to traditional docking methods such as Vina and Lin_F9. To this aim, we explore the utility of a data-driven platform called the minimum distance matrix representation (MDMR) to retrospectively predict recently discovered allosteric inhibitors complexed with Cyclin-Dependent Kinase (CDK) 2. In contrast to other protein complex representations, it uses the minimum residue–residue (or residue–ligand) distance as a feature that prioritizes the formation of interactions. Analysis of this representation highlights the variety of protein conformations and ligand binding modes, and we identify an intermediate protein conformation that other heuristic-based kinase conformation classification methods do not distinguish. Next, we design self- and cross-docking benchmarks to assess whether docking methods can predict both orthosteric and allosteric binding modes and if prospective success is conditional on the selection of the protein receptor conformation, respectively. We find that a combined method, DiffDock followed by Lin_F9 Local Re-Docking (DiffDock + LRD), can predict both orthosteric and allosteric binding modes, and the intermediate conformation must be selected to predict the allosteric pose. In summary, this work highlights the value of a data-driven method to explore protein conformations and ligand binding modes and outlines the challenges of SBDD of allosteric compounds.

In this work, a reliable and robust trust region method for restricted minimizations on hyperspheres is developed. The working equations of this new methodology are presented, together with their validation. The performance and characteristics of this new algorithm are discussed by a constrained minimization on a sphere using a two-dimensional Quapp model surface. The obtained results show that the proposed method for minimizations on hyperspheres guarantees convergence to constrained minima. Its enhanced numerical stability permits tight convergence criteria for constrained minimizations. The application of the new restricted minimizer in the framework of the hierarchical transition state finder and for the calculation of intrinsic reaction coordinates for 38 chemical reactions demonstrates its robustness and efficiency.