Assessing the performance of MM/PBSA and MM/GBSA methods. 8. Predicting binding free energies and poses of protein–RNA complexes

Fu Chen, Huiyong Sun, Junmei Wang, Feng Zhu, Hui Liu, Zhe Wang, Tailong Lei, Youyong Li, Tingjun HouRNA 2018 · DOI

Abstract

Accurately predicting the structural details and binding affinities of protein–RNA interactions (PRIs) is a cornerstone of computational biology, yet it remains notoriously difficult due to the flexibility and high charge density of RNA. This study, the eighth in a series evaluating molecular mechanics methods, benchmarks the performance of MM/PBSA and MM/GBSA scoring functions. By systematically exploring different solvent models and interior dielectric constants ($\epsilon_{in}$), the researchers demonstrate that MM/GBSA rescoring significantly outperforms traditional docking scoring functions. Specifically, the GB$^{GBn1}$ model with $\epsilon_{in} = 2$ achieved the highest correlation with experimental binding data and successfully identified near-native structural poses in nearly 80% of cases, offering a robust computational framework for studying RNA-related diseases.

Introduction: The Challenge of Protein–RNA Interactions

Protein–RNA interactions are essential to biological processes such as gene regulation and protein synthesis. However, modeling these interactions is computationally demanding. Standard molecular docking generates numerous potential binding structures (decoys), but the scoring functions used to rank them often fail for RNA. Unlike proteins, RNA molecules are highly flexible and carry a dense negative charge, which complicates the physics of binding. This research investigates whether physics-based methods like MM/PBSA and MM/GBSA can bridge the accuracy gap left by empirical docking scores.

Methods and Optimization

The study compared two primary physics-based approaches: Molecular Mechanics/Poisson–Boltzmann Surface Area (MM/PBSA) and Molecular Mechanics/Generalized Born Surface Area (MM/GBSA). These methods estimate binding free energy by combining gas-phase energy with continuum solvent models.

To optimize these methods for RNA, the researchers tested several variables:

  • Solvent Models: Implicit solvent approximations versus explicit TIP3P water molecules.
  • Dielectric Constants ($\epsilon_{in}$): Values ranging from 1 to 8.
  • Sampling Techniques: Simple energy minimization versus 1-nanosecond Molecular Dynamics (MD) simulations.

Datasets

The study utilized two distinct datasets to validate the methods:

  • Data Set I: 55 protein–RNA complexes with high-resolution crystal structures and experimental binding affinities ($pK_d$), used to test affinity prediction.
  • Data Set II: 148 non-redundant complexes with 3,600 decoys each, used to test the ability to identify native binding structures.

Key Insights: The Superiority of GB$^{GBn1}$ and $\epsilon_{in} = 2$

A critical finding was the "RNA dielectric shift." While an $\epsilon_{in}$ of 1 is standard for protein-protein studies, an $\epsilon_{in}$ of 2 proved superior for protein–RNA affinity. This suggests that the high charge density of RNA requires a more polarizable model to be accurate.

Furthermore, the GB$^{GBn1}$ model emerged as the most effective Generalized Born model for RNA systems. Interestingly, the study found that simple energy minimization in implicit solvent often provided a sufficient balance of speed and accuracy for high-throughput tasks.

Performance of various GB models across different dielectric constants.

Key Results: Affinity and Pose Recognition

The physics-based rescoring methods demonstrated a clear advantage over traditional empirical scoring functions:

  • Affinity Prediction: MM/GBSA (using the GB$^{GBn1}$ model and $\epsilon_{in}=2$) achieved a Pearson correlation ($r_p$) of -0.557 with experimental data. This is a substantial improvement over standard docking scores like DECK ($r_p = -0.034$) and DARS ($r_p = -0.414$).
  • Pose Recognition: In a benchmark of 148 complexes, MM/GBSA identified a near-native structure within the top 10 results for 79.1% of the cases.
Correlation of traditional docking scoring functions with experimental data.

Conclusion and Future Directions

MM/GBSA rescoring is an efficient way to improve the prediction capability of scoring functions for protein–RNA systems. While challenges remain—such as accounting for large-scale RNA conformational changes (flexible docking) and the inherent computational cost of physics-based methods—this study provides a clear benchmark for the field. As more protein-RNA structures are experimentally determined, these models will continue to be refined for drug discovery and biological research.

Summary updated

Abstract

Molecular docking provides a computationally efficient way to predict the atomic structural details of protein-RNA interactions (PRI), but accurate prediction of the three-dimensional structures and binding affinities for PRI is still notoriously difficult, partly due to the unreliability of the existing scoring functions for PRI. MM/PBSA and MM/GBSA are more theoretically rigorous than most scoring functions for protein-RNA docking, but their prediction performance for protein-RNA systems remains unclear. Here, we systemically evaluated the capability of MM/PBSA and MM/GBSA to predict the binding affinities and recognize the near-native binding structures for protein-RNA systems with different solvent models and interior dielectric constants (<i>ε</i><sub>in</sub>). For predicting the binding affinities, the predictions given by MM/GBSA based on the minimized structures in explicit solvent and the GB<sup>GBn1</sup> model with <i>ε</i><sub>in</sub> = 2 yielded the highest correlation with the experimental data. Moreover, the MM/GBSA calculations based on the minimized structures in implicit solvent and the GB<sup>GBn1</sup> model distinguished the near-native binding structures within the top 10 decoys for 117 out of the 148 protein-RNA systems (79.1%). This performance is better than all docking scoring functions studied here. Therefore, the MM/GBSA rescoring is an efficient way to improve the prediction capability of scoring functions for protein-RNA systems.
BibTeX
@inproceedings{chen2018assessing,
  author = {Fu Chen and Huiyong Sun and Junmei Wang and Feng Zhu and Hui Liu and Zhe Wang and Tailong Lei and Youyong Li and Tingjun Hou},
  title = {{Assessing the performance of MM/PBSA and MM/GBSA methods. 8. Predicting binding free energies and poses of protein–RNA complexes}},
  booktitle = {RNA},
  year = {2018},
  doi = {10.1261/rna.065896.118},
}
Figure