همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

AI-Driven Discovery of Novel AChE Inhibitors for Alzheimer's Disease Treatment: Model Development, Molecular Docking, and MD Simulation

نویسندگان :

Ahmad Ebadi¹ Fatemeh Noori²

1- دانشگاه علوم پزشکی همدان 2- دانشگاه علوم پزشکی همدان

کلمات کلیدی :

AI-assisted drug discovery،LSTM،Acetylcholinesterase inhibitors،Alzheimer’s disease،MD simulation

چکیده :

Background and Objective Alzheimer's disease (AD) is a progressive neurological disorder that primarily affects memory, thinking, and behavior (Stopschinski and Diamond, 2017). Acetylcholinesterase (AChE) is an essential enzyme in the nervous system responsible for breaking down acetylcholine, a neurotransmitter involved in transmitting signals between nerve cells (Chaves, Calheiro, 2020). In Alzheimer's, reduced levels of acetylcholine are linked to cognitive decline (Aktaş, Taslimi, Gülçin, Gök, 2017). The main goal of this study is to design a chemical language model (CLM) capable of generating novel AChE inhibitors that can effectively bind to AChE. Method & material A dataset of 613 known AChE inhibitors was compiled from the PubChem and ChemBL databases. The SMILES strings were tokenized into their constituent atomic and bonding elements. Data augmentation was employed to broaden the dataset, enabling the generation of a more diverse range of potential candidates. Here, 5 stages of data augmentation (0, 1, 3, 9, 19) were used. To convert each token into a dense vector representation, a 100-dimensional embedding layer was employed. The model architecture consisted of two LSTM layers, with 512 and 256 units, respectively. Training was carried out over 100 epochs. Libraries of compounds were then generated from the trained model at five distinct temperatures (0.2, 0.4, 0.6, 0.8, and 1), with each temperature yielding a unique set of compounds. A new library of novel and unique compounds was curated, and the highest-performing output was selected. Molecular docking and MD simulation were used to evaluate the generated compounds. MD simulations help to depict the long-range interactions, conformational changes, and flexibility of the inhibitor and protein. Findings The results obtained from this modeling were as follows: The model was trained using 9-fold data augmentation, resulting in more novel compounds at a temperature of 0.8. Overall, the model performed superbly in terms of considering compound novelty. At lower temperatures (0.2, 0.4), the model acted more conservatively, and generated more similar molecules. On the other hand, at higher temperatures (0.6, 0.8, 1.0), the model generated more diverse and novel molecules, possibly with new scaffolds or unconventional functional groups. The docking analysis demonstrated that the newly designed ligands, established crucial interactions with key tryptophan residues at positions 286 and 86. Additionally, the complex predicted by docking exhibited remarkable stability during a 100 ns molecular dynamics simulation. Conclusion The artificial intelligence model successfully learned the pharmacophore of AChE inhibitors from SMILES strings. After training, the model was able to generate novel inhibitors that retained the key pharmacophore features of the original compounds. Docking and molecular dynamics (MD) simulations further demonstrated that the predicted inhibitors effectively interact with crucial amino acid residues within the enzyme's active site.