Driven to near-experimental accuracy by refinement via molecular dynamics simulations

Abstract

Protein model refinement has been an essential part of successful protein structure prediction. Molecular dynamics simulation-based refinement methods have shown consistent improvement of protein models. There had been progress in the extent of refinement for a few years since the idea of ensemble averaging of sampled conformations emerged. There was little progress in CASP12 because conformational sampling was not sufficiently diverse due to harmonic restraints. During CASP13, a new refinement method was tested that achieved significant improvements over CASP12. The new method intended to address previous bottlenecks in the refinement problem by introducing new features. Flat-bottom harmonic restraints replaced harmonic restraints, sampling was performed iteratively, and a new scoring function and selection criteria were used. The new protocol expanded conformational sampling at reduced computational costs. In addition to overall improvements, some models were refined significantly to near-experimental accuracy.

Keywords: CASP; Markov-state modeling; model refinement; molecular dynamics simulation; protein structure prediction.

Figure 1.

Overall refinement protocol applied by FEIGLAB group in CASP13.

Figure 2.

Overall performance in CASP13 based on GDT-HA, Cα-RMSD, GDC-SC metrics. “Model 1” submissions (A–C), refined models only with the iterative protocol (D–F), and refined models only with the conservative protocol (G–I) are shown for comparison between each protocol. Solid diagonal lines and dashed diagonal lines with offsets (± 5 for GDT-HA and GDC-SC and ± 0.5 Å for Cα-RMSD) are shown to provide visual guidance.

Figure 3.

Refinement performance as a function of the initial model quality for the CASP13 targets (A) and benchmark targets (B). Boxplots show the statistics for each initial model quality bins. For the boxplots, boxes represent the first and the third quartiles, medians are marked with red horizontal lines, whiskers range between the minimum and the maximum data within 1.5 interquartile from the first and the third quartiles. Raw data points are overlaid as grey Xs. Linear regressions of raw data points are shown as black dashed lines, and the corresponding Pearson’s correlations coefficients are shown at the bottom of each box.

Figure 4.

Refinement examples for targets R0974s1 (A), R0986s1 (B), R1004-D2 (C), and R1002-D1 (D). Native structures, initial models, and refined models are depicted in yellow, sky blue, and magenta, respectively. Incorrect regions in the initial models that were significantly refined are indicated with blue arrows, while regions that could not be resolved are indicated with red arrows. Buried side-chains are also shown as sticks in (A and B). The structure quality before and after refinement in terms of GDT-HA CαRMSD, and GDC-SC is given below each model.

Figure 5.

Comparison of improvements in GDT-HA or RMSD values as a function of protein size in terms of the radius of gyration (A, B), initial model stability (C, D), and the number of clusters when simulated with the conservative protocol (E, F). Results for the iterative and conservative protocols are colored in blue and magenta, respectively. Benchmark results are shown as boxplots, and CASP13 results as scattered circles. For details about boxplots, see Figure 3. Outliers are indicated by Xs.

Figure 6.

Free energy landscapes for target R0974s1 from MSM analysis as a function of the first two tIC coordinates from unrestrained simulations (A), simulations according to the iterative protocol after each iteration (B) and according to the conservative protocol (C). Contour maps are shown for every 0.5 kcal/mol up to 6.0 kcal/mol. Projections of the native and the initial models are marked with blue and black Xs, respectively. States identified from MSM analysis are further characterized in terms of their populations (colored bars) and Cα-RMSD from the native structure (black line) (D). Population bars are colored as follows: blue - unrestrained simulation; green/yellow/orange - iterative protocol after first/second/third iteration; red - conservative protocol.