Could statistical potential models achieve comparable or better performance than deep learning models?
Briefings in Bioinformatics · 2026
DOI: 10.1093/bib/bbag088 ↗Kurz gefasst
Fachartikel in der Zeitschrift Briefings in Bioinformatics, 2026. Verfasst von Zhihao Wang und 6 weiteren Autor:innen. Beteiligte Einrichtung: School of Physics, Shandong University, 27 Shanda Nan Road, 250100 Jinan, Shandong Province, China. Hauptthemen laut MEDLINE: Deep Learning, Proteins, Models, Statistical, Molecular Docking Simulation. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 64 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Key R&D Program of China, National Natural Science Foundation of China, National Natural Science Foundation of Shandong Province.
Abstract (Originalsprache)
Accurately predicting protein–ligand interactions is vital for structure-based drug discovery. Although deep learning (DL) models have shown strong performance, the potential of traditional statistical potentials under data-limited conditions remains underexplored. Here, we systematically assess several statistical potential models in docking and virtual screening. We find that docking benefits from distance-dependent pairwise atom–atom potentials with clear physical meanings, while screening relies more on orientation-dependent atom–residue potentials that capture local chemical environments. Based on these findings, we propose HybridSP, a hybrid potential combining distance-dependent atom–atom, atom–residue, and orientation-dependent atom–residue terms. An affinity-weighted scheme is applied to correct biases in statistical distributions. On the CASF-2016 benchmark, HybridSP achieves a 91.6% docking success rate and an enrichment factor of 29.35 at the top 1%, rivaling and even surpassing state-of-the-art DL models. Its strong screening ability is further validated on directory of useful decoys-enhanced and directory of useful decoys-adjusted. These results demonstrate that well-designed statistical potentials can achieve high performance and interpretability without complex DL architectures, offering an efficient alternative for scoring function design. The models are available at: https://github.com/zelixirSH/HybridSP.git.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Substanzen
Autor:innen und Institutionen
- School of Physics, Shandong University, 27 Shanda Nan Road, 250100 Jinan, Shandong Province, China.
- Shanghai Zelixir Biotech Co. Ltd., 298 Xiangke Road, 201210 Shanghai, China.
- Centre for Artificial Intelligence Driven Drug Discovery, Faculty of Applied Science, Macao Polytechnic University, R. de Luís Gonzaga Gomes, 999078 Macao, China.
- School of Biological Sciences, Nanyang Technological University, 50 Nanyang Avenue, 639798, Singapore.
Förderung
- National Key R&D Program of China
- National Natural Science Foundation of China
- National Natural Science Foundation of Shandong Province
- Taishan Scholars Program of Shandong Province
- Science and Technology Development Fund of Macau
- Core Facility Sharing Platform of Shandong University
- National Demonstration Center for Experimental Physics Education (Shandong University)
Zitieren
APA
Wang, Z., Wang, S., Guo, J., Mu, Y., Liu, X., Zheng, L., & Li, W. (2026). Could statistical potential models achieve comparable or better performance than deep learning models?. Briefings in Bioinformatics. https://doi.org/10.1093/bib/bbag088
Vancouver
Wang Z, Wang S, Guo J, Mu Y, Liu X, Zheng L, et al. Could statistical potential models achieve comparable or better performance than deep learning models?. Briefings in Bioinformatics. 2026. doi:10.1093/bib/bbag088
BibTeX
@article{Wang2026,
author = {Wang, Zhihao and Wang, Sheng and Guo, Jingjing and Mu, Yuguang and Liu, Xiangdong and Zheng, Liangzhen and Li, Weifeng},
title = {Could statistical potential models achieve comparable or better performance than deep learning models?},
journal = {Briefings in Bioinformatics},
year = {2026},
doi = {10.1093/bib/bbag088}
}
EUVIMED