Integrative machine learning and molecular simulation approaches identify GSK3β inhibitors for neurodegenerative disease therapy

Abstract

Neurodegenerative diseases (NDDs), including Alzheimer's disease (AD) and Parkinson's disease (PD), are a growing global health concern, especially among the elderly, posing significant challenges to well-being and survival. GSK3β, a serine/threonine kinase, is a key molecular player in the pathogenesis of NDDs. Dysregulated activity of GSK3β has been linked to neurodegenerative complications. Targeting GSK3β with active-site-specific inhibitors presents a promising therapeutic strategy for mitigating its pathological effects and potentially intercepting NDD progression. This study aimed to identify potential GSK3β inhibitors through an integrated in silico approach combining machine learning (ML)-based virtual screening, molecular docking, molecular dynamics (MD) simulations, and MM/GBSA binding free energy calculations. ML models were trained using known GSK3β inhibitors from BindingDB. Among all models, the Random Forest (RF) algorithm had the best prediction accuracy, with a value of 0.6832 on the test set and 0.7432 on the training set, and was employed to screen the target library of 11,032 phytochemicals. The ML-based screening identified 2,898 compounds with potential inhibitory action against GSK3β. Further screening based on Lipinski's Rule of Five gave 221 drug-like candidates. These compounds were further evaluated for GSK3β interaction via molecular docking. The analyses found ZINC136900288, ZINC7267, and ZINC519549 bind strongly and interact well with key residues in GSK3β active site with their binding scores being - 9.9, -8.8, and - 8.7 kcal/mol, respectively. MD simulations were conducted for both ligand-bound and apo GSK3β to assess structural stability. The simulation results showed that the ligand bound complexes were structurally stable with less fluctuations and higher conformational stability. In addition, (MM/GBSA) binding free energy calculations were carried out to quantify the affinity of the candidate compounds, and the candidate compound ZINC136900288 has the strongest binding affinity (-24.86 kcal/mol) of the three. Notably, these identified compounds feature novel chemical scaffolds that are structurally distinct from previously reported GSK3β inhibitors, emphasizing the originality and therapeutic potential of this study. These results show that ZINC136900288 may serve as suitable GSK3β inhibitors. Nevertheless, the efficacy and safety of these compounds need to be further validated experimentally and further studied in vivo for possible therapeutic application in NDDs.

Keywords: Alzheimer’s’ disease; Glycogen synthase kinase-3 beta; Inhibitors; Machine learning; Molecular docking; Molecular dynamics simulation; Parkinson’s disease.

Fig. 1

(A) structure of GSK3β (PDB ID 4ACC) consisting of N-terminal β-sheet domain and C-terminal α-Helix domain. The hinge region (red), Activation loop (blue), and glycine-rich region (green). (B) The close-up representation of the important catalytic residues is involved in interactions. (C) Role of GSK3β in Alzheimer’s disease (AD). Upregulation of the GSK3β due to upstream regulators activates the β-catenin phosphorylation, neuronal dysfunctions, and Tau Hyperphosphorylation; hallmarks of AD. (D) Role of GSK3β in Parkinson’s disease. Upregulation of GSK3β leads to oxidative stress, neuroinflammation, and mitochondrial dysfunction.

Fig. 2

PCA Scatter Plot (PC1 vs. PC2) illustrating the distribution of compounds based on their principal components. The two classes are represented as Label 0 (blue) and Label 1 (green), indicating different categories of phytochemical compounds.

Fig. 3

Chemical space distribution of the training (A) and test (B) sets based on molecular weight (MW) and LogP. Color code of each point is blue for active compound and green for inactive compounds.

Fig. 4

ROC curves for different machine learning models evaluated on the training and test sets.

Fig. 5

Docked conformation of the three selected ligands in the active site of GSK3β (ZINC136900288 (yellow), ZINC7267 (blue), and ZINC519549 (pink).

Fig. 6

Interactions of the three ligands with the active site of GSK3β. (A) Reference, (B) ZINC136900288, (C) ZINC7267, and (D) ZINC519549. Residues making hydrogen bonds are shown as sticks and labeled. Residues making other interactions are shown as lines.

Fig. 7

(A) Root mean square deviations and (B) Root mean square fluctuations plotted over the time of 100ns. (C) The areas with high RMSF are shown as red and labeled.

Fig. 8

Ligand-GSK3b complexes posed extracted after 100ns simulation superimposed to docked complexes. The docked complex is represented in green color and simulated complex is represented in pink color.

Fig. 9

Solvent accessible surface area (SASA) and Radius of gyration of all the simulated systems plotted over the time of 100ns.

Fig. 10

Principal component analysis (PCA) of Apo and ligand-bound complexes.

Fig. 11

Physicochemical Properties comparison between the selected compounds (ZINC136900288, ZINC7267, and ZINC519549) and the known inhibitors Kenpaullone, Tideglusib, and SB216763.