EUVIMEDEUVIMED entdecken
Crossref

AI-driven protein pocket detection through integrating deep Q-networks for structural analysis

Prashanth Choppara, Lokesh Bommareddy

Journal of Computer-Aided Molecular Design · 2025

DOI: 10.1007/s10822-025-00669-2 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Computer-Aided Molecular Design, 2025. Verfasst von Prashanth Choppara und 1 weiteren Autor:innen. Beteiligte Einrichtung: SCOPE, VIT-AP University, Amaravathi, Andhra Pradesh, 522237, India. Hauptthemen laut MEDLINE: Proteins, Deep Learning. Die Arbeit nennt 37 Literaturangaben.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Protein pockets, or small cavities on the protein surface, are critical sites for enzymatic catalysis, molecular recognition, and drug binding. Accurately identifying these pockets is crucial for understanding protein function and designing therapeutic interventions. Traditional computational methods such as molecular docking, surface grid mapping, and molecular dynamics simulations are hampered by the use of fixed protein structures, and therefore it is challenging to identify cryptic pockets when they appear under physiological conditions. We propose a deep reinforcement learning (DRL) technique based on deep Q-networks (DQN) to identify precise protein pockets. Our strategy to improve the prediction of functional binding sites incorporates important molecular descriptors such as spatial coordinates, solvent-accessible surface area (SASA), hydrophobicity, and electrostatic charge. We pre-process protein structure data from the protein data bank (PDB) through feature extraction and selection methods, including variance threshold filtering and dimensionality reduction using an autoencoder. The sparse feature representation enables efficient training of a DQN agent, which navigates protein surfaces and iteratively optimizes pocket predictions. By using reinforcement learning concepts, the model adapts its pocket detection strategy according to the learned reward signals, increasing sensitivity and specificity. The method is tested on benchmark datasets and is found to exhibit superior performance in detecting well-defined and cryptic pockets over traditional computational methods. Experimental evidence suggests that our model successfully identifies binding sites in various protein families, with significant implications for drug discovery and protein-ligand interaction studies. Moreover, the model's ability to incorporate geometric and biochemical features allows for a better understanding of pocket functionality. The scalability of our method makes it an important tool for large-scale virtual screening and personalized medicine. By using deep reinforcement learning, this research provides a new and effective framework for protein pocket prediction, opening up opportunities for developing new tools in structural bioinformatics, drug design, and molecular biology research.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

ProteinsDeep LearningBinding SitesMolecular Docking SimulationDatabases, ProteinProtein ConformationMolecular Dynamics SimulationHumansProtein BindingAlgorithms

Stichwörter der Autor:innen

Deep Q-network (DQN)Dimensionality reductionFeature extractionHydrophobicityMolecular dockingProtein pocketsSolvent-accessible surface area (SASA)Structural bioinformatics

Substanzen

Proteins

Autor:innen und Institutionen

  • SCOPE, VIT-AP University, Amaravathi, Andhra Pradesh, 522237, India.
  • SCOPE, VIT-AP University, Amaravathi, Andhra Pradesh, 522237, India. lokesh.bommareddy@vitap.ac.in.

Zitieren

APA

Choppara, P., & Bommareddy, L. (2025). AI-driven protein pocket detection through integrating deep Q-networks for structural analysis. Journal of Computer-Aided Molecular Design. https://doi.org/10.1007/s10822-025-00669-2

Vancouver

Choppara P, Bommareddy L. AI-driven protein pocket detection through integrating deep Q-networks for structural analysis. Journal of Computer-Aided Molecular Design. 2025. doi:10.1007/s10822-025-00669-2

BibTeX
@article{Choppara2025,
  author = {Choppara, Prashanth and Bommareddy, Lokesh},
  title = {AI-driven protein pocket detection through integrating deep Q-networks for structural analysis},
  journal = {Journal of Computer-Aided Molecular Design},
  year = {2025},
  doi = {10.1007/s10822-025-00669-2}
}
Kostenlos starten →