Physical determinants of reliable RNA–ligand binding prediction revealed by P-RISE
- 1Xiangya School of Pharmaceutical Sciences, Central South University , Changsha 410004 ,
Abstract
RNA-targeted small-molecule discovery is challenged by the scarcity of experimentally characterized RNA-ligand complexes and by the limited transferability of current computational approaches across chemically diverse systems. Here we present P-RISE, a physics-grounded framework for RNA-ligand binding prediction, and use it to identify the physical determinants underlying reliable performance. Systematic evaluation of force fields, ionic environments, dielectric treatments, and entropy corrections reveals three coupled requirements: preservation of RNA-binding pocket integrity under ion-sensitive conditions; an effective electrostatic error cancellation regime enabled by an elevated internal dielectric constant; and an empirically observed coupling between entropy correction and solvent modeling. Integrating these features, P-RISE provides stable affinity ranking, physics-grounded delineation of its physicochemical applicability domain, accurate metal ion site localization, and improved binding pose discrimination relative to established RNA docking and scoring approaches. Crucially, benchmarking against AlphaFold3 reveals that co-folding models, optimized for structure generation rather than energetic ranking, benefit from an explicit physics-based evaluation layer. Furthermore, diverse machine-learning comparisons demonstrate that purely data-driven methods lack stability and generalization in this data-poor regime. Ultimately, P-RISE provides a mechanistic understanding of RNA-ligand modeling, serving as a principled physical baseline that defines reliable prediction boundaries, and guides future data-driven integration.
Related
- Representation-aware conformal prediction for reliable uncertainty quantification in drug-target affinity
- The extraordinary robustness of mitotic spindle assembly to microtubule dynamics plasticity revealed by end-binding proteins and tubulin tagging
- An optimized machine learning approach for reliable agronomic parameter prediction in precision farming systems
- A method for brain tumor detection and localization by means of deep learning and reliable prediction explainability
- Benchmarking Large Language Models for Assay-Aware Drug-Target Binding Affinity Prediction
BibTeX
@article{li2026physical,
title = {{Physical determinants of reliable RNA–ligand binding prediction revealed by P-RISE}},
author = {Hui Li and Youchao Deng and Zhuoxi Jiang and Xinyu Gu and Dejun Jiang and Dongsheng Cao},
journal = {Nucleic Acids Research},
year = {2026},
month = oct,
doi = {10.1093/nar/gkag1062},
url = {https://academic.oup.com/nar/article/doi/10.1093/nar/gkag1062/9006198},
note = {CC BY},
}
RIS
TY - JOUR
TI - Physical determinants of reliable RNA–ligand binding prediction revealed by P-RISE
AU - Hui Li
AU - Youchao Deng
AU - Zhuoxi Jiang
AU - Xinyu Gu
AU - Dejun Jiang
AU - Dongsheng Cao
PY - 2026
DA - 2026/10/01
JO - Nucleic Acids Research
DO - 10.1093/nar/gkag1062
UR - https://academic.oup.com/nar/article/doi/10.1093/nar/gkag1062/9006198
AB - RNA-targeted small-molecule discovery is challenged by the scarcity of experimentally characterized RNA-ligand complexes and by the limited transferability of current computational approaches across chemically diverse systems. Here we present P-RISE, a physics-grounded framework for RNA-ligand binding prediction, and use it to identify the physical determinants underlying reliable performance. Systematic evaluation of force fields, ionic environments, dielectric treatments, and entropy corrections reveals three coupled requirements: preservation of RNA-binding pocket integrity under ion-sensitive conditions; an effective electrostatic error cancellation regime enabled by an elevated internal dielectric constant; and an empirically observed coupling between entropy correction and solvent modeling. Integrating these features, P-RISE provides stable affinity ranking, physics-grounded delineation of its physicochemical applicability domain, accurate metal ion site localization, and improved binding pose discrimination relative to established RNA docking and scoring approaches. Crucially, benchmarking against AlphaFold3 reveals that co-folding models, optimized for structure generation rather than energetic ranking, benefit from an explicit physics-based evaluation layer. Furthermore, diverse machine-learning comparisons demonstrate that purely data-driven methods lack stability and generalization in this data-poor regime. Ultimately, P-RISE provides a mechanistic understanding of RNA-ligand modeling, serving as a principled physical baseline that defines reliable prediction boundaries, and guides future data-driven integration.
SN - 0305-1048
SN - 1362-4962
ER -