Journal of Computer-Aided Molecular Design最新文献

The discovery of dual acetylcholinesterase (AChE) and β-secretase (BACE1) inhibitors remains a promising strategy against multifactorial Alzheimer’s disease. Here, rigorously curated ChEMBL-derived data were used to develop explainable QSAR (Quantitative structure–activity relationship) models for dual-inhibition prioritization. Molecules were standardized, near-duplicates were removed using a Tanimoto similarity threshold (≥ 0.80), and physicochemical outliers were filtered prior to modeling. Multiple classifiers (including Light Gradient-Boosting Machine, eXtreme Gradient Boosting, Random Forest, Support Vector Machine, k-Nearest Neighbors and Gradient Boosting Decision Trees) and fingerprints (e.g., RDKit fingerprints, Extended Connectivity Fingerprint) were benchmarked under scaffold-based nested cross-validation to prevent data leakage. Class imbalance was handled with SMOTETomek applied strictly within training folds. Model selection relied on F-Score, Area Under the Precision–Recall Curve, Matthews Correlation Coefficient (MCC), and Recall, and performance was accompanied by bootstrap confidence intervals, calibration curves, and Y-randomization controls. In classification, the top model (GBDT + ECFP6) achieved strong generalization (Recall ≈ 1.00, PR-AUC ≈ 0.84, MCC ≈ 0.81, F1 Score ≈ 0.84). Shapley Additive Explanations (SHAP) analysis highlighted aromatic and hydrogen-bonding substructures as key positive contributors. Prospective candidates (e.g., CHEMBL5082250, CHEMBL1651126, CHEMBL1651127) were evaluated by active-site-focused docking against AChE (PDB: 4EY7) and BACE1 (PDB: 2G94) with essential waters retained; docking scores (ΔG, kcal·mol⁻¹) were used for relative ranking of the ligands. SwissADME/pkCSM profiling suggested CNS-relevant properties (e.g., MPO, logBB, P-gp liability) and acceptable oral drug-likeness. Collectively, the workflow provides a reproducible and transparent pipeline for prioritizing dual AChE/BACE1 chemotypes and nominates testable scaffolds for experimental validation.

Inhibition of Coagulation Factor X (FXa) is a clinically validated therapeutic strategy; however, developing safer and more selective inhibitors remains a major challenge. In this study, we developed an interpretable machine learning–based QSAR framework to predict both the inhibitory potency and activity class of small molecules targeting FXa. A structurally curated dataset of 6400 compounds was retrieved from ChEMBL, standardized, and encoded using 391 non-redundant Mordred descriptors following systematic filtering. Benchmarking of 42 regression and 42 classification algorithms identified ExtraTreesRegressor and XGBoostClassifier as the most robust models. The regression model achieved an R² of 0.760 and an RMSE of 0.831 on the independent test set, while the classification model reached an accuracy of 0.91 with balanced precision, recall, and an ROC-AUC of 0.962. SHAP (SHapley Additive exPlanations) analysis further enhanced interpretability by revealing that electrostatic, topological, and polar surface descriptors were the dominant contributors to FXa inhibitory potency. Applicability domain assessment using Williams plots confirmed that most compounds in both the training and test sets lay within the model’s reliable prediction space. Overall, the proposed QSAR pipeline integrates strong predictive performance with valuable mechanistic interpretability and rigorous validation, offering a practical computational tool for the virtual screening and rational design of novel FXa inhibitors.

Survivin, a multifunctional regulator of mitosis and apoptosis, plays a central role in cancer progression and therapy resistance, making it an attractive target for anticancer drug development. In this study, a series of 4,6-diphenyl-3-cyanopyridine derivatives were designed and synthesized as potential survivin inhibitors through an integrated strategy combining 3D-QSAR modeling, molecular docking, molecular dynamics simulations, and biological evaluation. The CoMFA and CoMSIA models established reliable structure–activity relationships and provided contour-map-based guidance for rational molecular optimization. Newly designed derivatives displayed enhanced antiproliferative effects against melanoma cells, and computational analyses revealed that the most promising compound showed stable and preferential binding within the BIR domain of survivin, particularly in its dimeric form. These findings demonstrate the effectiveness of contour-guided optimization in discovering novel survivin-targeting scaffolds and highlight 4,6-diphenyl-3-cyanopyridine derivatives as promising leads for further anticancer drug development. Future studies will focus on improving selectivity, clarifying the inhibition mechanism at the molecular level, and evaluating in vivo efficacy.

Flavonoids are found in most edible plants and vegetables and due to their specialized chemical structures and biological activities, we aimed to investigate the efficacy of selected common flavonoids against diabetes-related advance glycation end products (AGEs) through both computational and experimental approaches. Major in silico techniques involved network pharmacology, molecular docking and in vitro AGEs inhibition assays. The pathway enrichment analysis revealed a significant association between AGE regulation and several key biological pathways, including those involved in phenylalanine metabolism, Th17 cell differentiation, and sphingolipid signaling. Molecular docking revealed that hesperidin exhibited the highest binding affinities with transcription regulators 3CJJ (ΔG − 7.1 kJ/mol) and 3TOP (ΔG − 10.0 kJ/mol), while epicatechin showed strong binding to 4F5S (ΔG − 8.3 kJ/mol). All tested compounds significantly reduced oxidative stress, with hesperidin demonstrating moderate inhibition of advanced glycation in the bovine serum albumin (BSA)-glucose model (61.2% ± 1.4%) and BSA-MGO model (52.1% ± 1.7%), as well as potent α-glucosidase inhibition (IC₅₀ = 22.43 ± 1.84 µM). Mechanistic studies further showed moderate protective effects against β-amyloid aggregation and effective trapping of fructosamine and carbonyl groups. The findings suggest that hesperidin and epicatechin possess strong anti-AGEs and anti-inflammatory activities.

Botulinum neurotoxins are class I tier bioterrorism agent, accountable for causing rare but fatal illness ‘botulism’. Out of seven serotypes, A, B, E, and F are responsible for intoxicating humans. Despite knowing harmful effects on human health for centuries, there is no commercial antidote for post-neuronal intoxication is available. In the present study, we report efficacy of regioisomers of [(8-hydroxyquinolin-7-yl)(phenyl) methylamino]benzoic acid against zinc-dependent light chain activities of BoNT/A, /B, /E & /F by combining molecular modeling with in vitro and in vivo studies. Based on structure similarity search, multiple regioisomers of 8-hydroxyquinoline were mined and screened by performing molecular docking. The best-scored compounds were analyzed for inhibitory and binding potential against these serotypes via endopeptidase and surface plasmon resonance assays. The best two compounds (NSC1011 and NSC1012) with potential inhibition and binding kinetics across serotypes were evaluated for therapeutic potential in mouse model. NSC1011 and NSC1012 (regioisomers) docking data revealed their binding energies with active domains of BoNT/A, /B, /E, and /F light chains ranging between − 9.70 to − 4.27, and  − 9.84 to − 7.23 kcal/mol, respectively. The endopeptidase assay displayed ˃ 90% inhibition of catalytic activities, with the IC₅₀ values varying among serotypes from 20 to 40 µM concentrations. SPR interaction of both compounds with the targeted proteins was observed in the range of 3.83E-05 to 4.95E-04 M. These molecules have shown complete protection at one MLD (mouse lethal dose), whereas median extension of animal survival was recorded up to 24 h when exposed to 5X MLD. The in silico, in vitro, and in vivo data reveal that NSC1011 and NSC1012 exhibited good binding affinity, stability, inhibition with promising therapeutic potential against human botulism-causing toxinotypes.

We introduce a generative drug-design framework that combines large chemical language models (CLMs) pretraining, target specific masked-language fine-tuning, and reinforcement learning (RL) to create novel small molecule inhibitors of EGFR. Using a multi-objective reward that balances predicted potency, drug-likeness, synthetic accessibility, and structural novelty, the model learns to explore chemically valid and diverse regions of EGFR-relevant chemical space beyond known inhibitors. The resulting compounds exhibit improved computational binding trends relative to reference EGFR inhibitors and include highly novel chemotypes with no close analogs in the training set. This study demonstrates how integrating pretrained chemical language models with reinforcement learning can accelerate target focused de novo molecular design and provides a generalizable framework for future applications in kinase inhibitor discovery.

Type II diabetes mellitus is a major endocrine disorder characterized by persistent hyperglycemia, insulin resistance, and dysregulation in glucose uptake by the cells. Peroxisome proliferator-activated receptor-γ (PPARγ) plays a significant role in the regulation of glucose and lipid metabolism as well as in post-diabetic inflammatory response. Therefore, PPARγ activators seem to be the drugs of choice. In the present work, structure-based virtual screening approach was employed to find newer compounds as PPARγ agonist. The ChemDiv library (freely available) of compounds was used for hierarchical virtual screening; the hits obtained were further evaluated based on in silico predicted binding energy and toxicity predictions. The structure-based approach yielded 18 high-affinity, stably binding hits, from which 08 hits (Sn1-Sn8) were predicted to be non-toxic. Further, in vitro exploration of the anti-diabetic as well as PPARγ agonistic potential was carried out on eight (08) ligands obtained from in silico scrutiny, using various in vitro assays. The synthesized quinazolinedione based compound (Sn9) was also evaluated similarly for exploration of its lead-likeness as PPARγ agonistic anti-diabetic candidate. Compounds Sn7 and Sn8 showed adequate glucose uptake by the cells, anti-adipogenicity, and PPARγ binding, while Sn4 and Sn9 showed moderate potential in the same examination. Safety profiles of these compounds on 3T3-L1 and C2C12 cells were also established. The in vitro studies suggested that imidazopyridine (present in Sn4, Sn8) and quinazolinedione (present in Sn7 and Sn9) have much potential against T2DM. Sn8 was found to be the best candidate, and it also demonstrated a stable trajectory and interaction profile in simulated physiological environment. The study confirms the lead-like potential of compound Sn8, and supports the exploration of imidazopyridine and quinazolinedione ring systems for further development of PPARγ agonistic lead compounds in the anti-diabetic arena.

Graphical abstract

The integration of artificial intelligence (AI) with molecular modeling offers new opportunities to accelerate therapeutic discovery. In this study, we developed an AI-driven generative pipeline combining deep reinforcement learning (DRL), generative adversarial networks (GANs), and variational autoencoders (VAEs) to design novel peptide-like molecules targeting major proteins implicated in endometrial cancer (EC), including AKT1, ESR1, Connexin-43, and CTNNB1. From over 14,200 generated structures, approximately 2313 peptides met drug-likeness and structural criteria and were screened using deep learning-enhanced docking. Top-ranked peptides, such as Gitoxoside (− 11.53 kcal/mol) and 9-Fluoro-11 (− 11.38 kcal/mol), demonstrated stronger binding to AKT1 than the reference inhibitor Capivasertib (− 8.50 kcal/mol). Similar high-affinity interactions were observed for CTNNB1–SCHEMBL (− 12.33 kcal/mol) and ESR1–1Estra-1,3 (− 11.05 kcal/mol). Molecular dynamics (MD) simulations confirmed the stability of these complexes with RMSD values below 2.5 Å and minimal residue fluctuations. WaterSwap free energy calculations yielded highly favorable binding energies (− 34 to − 37 kcal/mol), further validating stable ligand–protein interactions. ADMET predictions indicated acceptable pharmacokinetic properties and low predicted toxicity for most candidates. Collectively, this integrative AI framework efficiently explores peptide chemical space, enabling the rapid identification of peptide-based and peptidomimetic inhibitors with strong binding affinity and stability. The findings highlight the potential of AI-assisted peptide design as a scalable and cost-effective strategy for developing next-generation therapeutics against endometrial cancer.

The c-Myc oncogene is crucial in tumorigenesis. Although it is a promising therapeutic target, its protein lacks a conventional drug-binding pocket, making it traditionally “undruggable”. Recent studies show that the c-Myc promoter can form a G-quadruplex (G4) structure, which suppresses transcription and offers a new strategy for indirect inhibition. In this study, structure-based virtual screening was performed using the c-Myc G4 crystal structure to screen the ChemDiv compound library, aiming to identify small molecules that bind to the G4 structure. Candidate compounds were evaluated in preliminary in vitro assays for biological activity. The results showed that Y502-3888 binds to the c-Myc G4 and downregulates c-Myc expression at both mRNA and protein levels. Collectively, these findings support the potential of Y502-3888 as a c-Myc G4 binder for the treatment of multiple myeloma (MM), providing a foundation for future development of anticancer agents targeting the c-Myc G4.

Graphical abstract

A prodrug is a pharmacologically inactive (or attenuated) derivative that undergoes bioreversible transformation in vivo to release an active parent drug, enabling temporary optimization of properties such as solubility, permeability, and targeting. Despite expanding catalogs of known prodrugs, in silico screening remains limited by the absence of reliable negative examples: training/evaluation sets often contain only positives or ad-hoc decoys, leading to class imbalance, property-mismatch shortcuts, and irreproducible benchmarks. Unfortunately, the limitation of reliable negatives has resulted in there being no efficient machine learning-based prodrug screening approach. Therefore, we introduce Prodrug-ML, an efficient machine learning-based screen for prodrug-likeness that prioritizes candidates rather than asserting mechanistic truth. Prodrug-ML helps medicinal chemists triage prodrugging ideas during hit-to-lead and lead optimization, filter enumerated libraries of promoiety–attachment variants before ADMET assays, and retrospectively mine internal/ChEMBL-like collections to surface likely prodrug chemotypes. In practice, users (i) generate or collect candidate structures (e.g., parent drug ± pro-moieties), (ii) score them with Prodrug-ML, and (iii) advance only high-scoring candidates to synthesis/assay, thereby reducing wet-lab load while maintaining chemical diversity. In order to achieve such practical usage, the Prodrug-ML framework, containing the default classifier, LightGBM, addresses these issues by (i) constructing three complementary, property-controlled negative cohorts (DUD-E–style near-misses, random ChEMBL, and strictly filtered ChEMBL), (ii) hardness control and label-noise guardrails on decoys, (iii) domain-bias control, and (iv) cross-decoy validation with multimodel feature selection. Produg-ML has been evaluated five times on hold-out data and an unseen test benchmark, after 80% of training data. In the benchmarks, the multimodel ensemble consistently improves early retrieval and overall discrimination, attaining (textrm{EF}@1%approx 6text {--}8), (textrm{EF}@5%approx 5text {--}6), (textrm{BEDROC}_{20}approx 0.78text {--}0.82), (textrm{BEDROC}_{50}approx 0.90text {--}0.95), and (textrm{BEDROC}_{80}approx 0.95text {--}0.99), alongside ROC AUC (approx 0.86text {--}0.87), average precision (approx 0.60text {--}0.65), and F1 (approx 0.58text {--}0.62). As a result, these results, especially high BEDROC scores, are consistent with concentrating at least a prodrug within the top (sim 2text {--}3%) of ranked candidates, implying (sim 97text {--}98%) reductions in experimental time and cost when using standard wet-lab workflows that assay only the early tranche.