Deep learning-guided de novo design and molecular dynamics validation of putative glucocorticoid receptor binders
Authors: Zheng, Junyuan; Zhang, Man; Huang, Biyun; Weng, Rongsheng; Yi, Wei; Zhao, Xin
Source journal: Journal of molecular modeling (2026-10-03)
DOI: 10.1007/s00894-026-06949-4
Abstract
CONTEXT: Glucocorticoid receptor (GR) ligands are widely used anti-inflammatory agents, but their therapeutic utility is limited by severe adverse effects associated with long-term administration, including metabolic disorders, osteoporosis, and suppression of the hypothalamic-pituitary-adrenal (HPA) axis. The discovery of structurally novel GR binders with improved safety profiles therefore remains a high-priority goal in drug development. In this purely computational study, a deep learning-guided de novo design framework based on maximum-likelihood training and transfer learning was developed to construct a focused library of 1,322 GR-relevant molecular entities. These molecules were initially screened using an ECFP4-based random forest (RF) classifier optimized for predicting GR activity, followed by docking-based virtual screening against the GR ligand-binding domain (LBD; PDB 4P6W). Five candidates were then prioritized on the basis of their binding profiles; however, their computed binding affinities remained substantially lower than that of the reference agonist mometasone furoate (MOF).Molecular dynamics (MD) simulations (500 ns x 3 independent replicates) were performed to evaluate complex stability, and MM/GBSA binding free-energy calculations further corroborated the binding profile. In silico ADME/toxicity profiling indicated favorable drug-like properties for the prioritized candidates. Overall, among the generated compounds, CPD_263 and CPD_313 achieve the best balance between predicted binding strength and safety-related parameters, emerging as computationally prioritized putative GR binders that merit subsequent experimental validation. All findings are computational and are intended to provide testable hypotheses for future experimental validation. METHODS: Molecular generation was performed using a recurrent neural network trained by maximum-likelihood pretraining followed by target-specific transfer learning. Activity prediction employed an ECFP4-based random forest classifier. Molecular docking was implemented with AutoDock Vina, and MD simulations were performed using GROMACS 2022.5 to assess the dynamic stability of the hit candidates. Binding free energies were calculated with MM/GBSA, and in silico ADMET predictions were performed using SwissADME and ProTox-3.0.
Related topics
Related open access articles
- Development and Validation Molecular Docking Analysis of Human serum albumin (HSA)
- Facing COVID-19 via anti-inflammatory mechanism of action: Molecular docking and Pharmacokinetic studies of six anti-inflammatory compounds derived from Passiflora edulis
- Computational analysis by molecular docking of thirty alkaloid compounds from medicinal plants as potent inhibitors of SARS-CoV-2 main protease
- Curcumin and pipernonaline to curb diabetes the natural way: A molecular modeling, docking and dynamics simulation study
- DENSITY FUNCTIONAL THEORY ANALYSIS AND MOLECULAR DOCKING EVALUATION OF 8-CHLOROQUINOLINE 2-CARBALDEHYDE
- Thermochemical, DFT, Molecular Docking and Pharmacokinetic Studies of Methyl β-D-galactopyranoside Esters
Abstract and bibliographic data from PubMed (U.S. National Library of Medicine). Copyright of the abstract belongs to the original publisher or authors.