سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

صفحه اصلی / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Enhancing Drug–Target Affinity Prediction with Non-Local Block Graph Neural Networks

نویسندگان :

Reza Tahmasebi¹ Eghbal Mansoori² Armin Piashehvar³ Abbas Mehrbaniyan⁴

1- دانشگاه شیراز 2- دانشگاه شیراز 3- دانشگاه شیراز 4- دانشگاه شیراز

کلمات کلیدی :

Drug target affinity،Graph neural network،Non-local attention،Protein ligand interaction،Virtual screening،Deep learning،Contact maps،Kinase

چکیده :

Predicting how strongly a drug binds to its protein target is key to speeding up drug discovery. Many graph neural networks excel with local chemical features but often overlook important long-range contacts, particularly when they become deep and challenging to train. In this paper, we present a compact graph neural network model, called NLB-DTA, that adds a single non-local attention block between two light convolutional layers in distinct ligand and protein streams. This block considers the contribution of each atom to other atoms in the molecular downstream and residues in the protein downstream in a single step, capturing distant interactions between atoms and residues. Tested on the widely used Davis dataset, the model achieves state-of-the-art accuracy with high generalization. Since NLB-DTA needs only standard ligand strings and protein sequences, it can run on common GPUs. Our study shows that using a non-local layer is effective in considering the long-range information gap, and so it offers a practical foundation for future virtual screening.