A Hybrid Metaheuristic for Molecular Docking

• Osmar Aguiar Ribeiro Jr Universidade Federal do Maranhão (UFMA)
• Omar Andres Carmona Cortes Instituto Federal de Educação, Ciência e Tecnologia do Maranhão (IFMA) / Universidade Federal do Maranhão (UFMA)
• João Otávio Bandeira Diniz Instituto Federal de Educação, Ciência e Tecnologia do Maranhão (IFMA) / Universidade Federal do Maranhão (UFMA)

DOI: https://doi.org/10.5753/bsb.2025.13645

Resumo

This paper presents a hybrid metaheuristic algorithm, ABC-GA-VGOS, integrating Artificial Bee Colony (ABC), Genetic Algorithm (GA), and Variable Genetic Operator Search (VGOS) for molecular docking. The algorithm initializes a population of size N , evaluates fitness using a heuristic docking energy (k-d tree with distance-based penalties), and employs adaptive crossover and mutation probabilities (pc, pm) to balance exploration and exploitation. Elitism preserves the best individual, while Solis-Wets’ local search refines solutions only when improvements are detected, optimizing computational efficiency. Experimental results on ten protein–peptide systems (1JSU, 1A1N, 1BE9, 1BXO, 1BXL, 1YCR, 2AN6, 2BBA, 2IGF, 4QVE) confirmed the robustness of the proposed method. In simpler systems such as 1JSU and 1A1N, all algorithms achieved 100% success rates (RMSD ≤ 2.0 Å) with mean RMSDs of 0.63 Å and 0.57 Å, respectively. In intermediate cases (e.g., 1BE9, 2AN6, 2BBA, 2IGF, 4QVE), the hybrid algorithm consistently outperformed the baselines, yielding more negative binding energies, lower RMSDs, and higher success rates. For highly complex systems (1BXL, 1BXO, 1YCR), none of the methods succeeded (RMSD > 3.5 Å), underscoring the challenges of flexible peptide docking. These results highlight the reliability of the proposed hybrid, making it a robust tool for protein–peptide docking, with potential for further optimization via advanced operator selection or differentiable energy models.

Palavras-chave: Molecular Docking

Referências

Leonhart, P. F. and Dorn, M. (2019). A biased random key genetic algorithm with local search chains for molecular docking. In Kaufmann, P. and Castillo, P., editors, International Conference on the Applications of Evolutionary Computation (EvoApplications), volume 11454 of Lecture Notes in Computer Science, pages 360–376. Springer.

Lin, C.-H. et al. (2025). Deeprli: a deep reinforcement learning-inspired graph neural network for multi-task protein–ligand docking. Digital Discovery, 4:403–417.

Masoudi-Sobhanzadeh, Y., Jafari, B., Parvizpour, S., Pourseif, M. M., and Omidi, Y. (2021). A novel multi-objective metaheuristic algorithm for protein–peptide docking and benchmarking on the leads-pep dataset. Computers in Biology and Medicine, 138:104896.

McNutt, A. T. et al. (2025). Gnina 1.3: molecular docking with deep learning-based scoring and covalent docking support. Journal of Cheminformatics, 17(1):97.

Morehead, A. et al. (2024). Posebench: a large-scale benchmark for evaluating deep learning-based protein–ligand docking methods. Nature Methods, 21:1234–1246.

Shirali, A., Alghamdi, A. A., et al. (2025). A comprehensive survey of deep learning vs classical scoring functions for molecular docking. Journal of Cheminformatics, 17(1):73.

Sob, M. et al. (2024). Reinforcement learning fine-tuning of generative latent space models improves molecular docking hit rates. arXiv preprint.

Tavares, J., Mesmoudi, S., and Talbi, E.-G. (2009). On the efficiency of local search methods for the molecular docking problem. In Pizzuti, C., Ritchie, M., and Giacobini, M., editors, EvoBIO 2009: European Conference on Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics, volume 5483 of Lecture Notes in Computer Science, pages 104–115. Springer.

Zhou, J., Yang, Z., He, Y., Ji, J., Lin, Q., and Li, J. (2023). A novel molecular docking program based on a multi-swarm competitive algorithm. Swarm and Evolutionary Computation, 78:101292.