PIE-efficient filters and coarse grained potentials for unbound protein-protein docking

Abstract

Identifying correct binding modes in a large set of models is an important step in protein-protein docking. We identified protein docking filter based on overlap area that significantly reduces the number of candidate structures that require detailed examination. We also developed potentials based on residue contacts and overlap areas using a comprehensive learning set of 640 two-chain protein complexes with mathematical programming. Our potential showed substantially better recognition capacity compared to other publicly accessible protein docking potentials in discriminating between native and nonnative binding modes on a large test set of 84 complexes independent of our training set. We were able to rank a near-native model on the top in 43 cases and within top 10 in 51 cases. We also report an atomic potential that ranks a near-native model on the top in 46 cases and within top 10 in 58 cases. Our filter+potential is well suited for selecting a small set of models to be refined to atomic resolution.

Figure 1

Sketch of the algorithm for efficient computation of surface area - (P, θ) overlap (P is the point where the line joining the centers of the spheres intersects surface of the sphere, and θ is the angle subtended at the center) (a), exposed surface in receptor (b), exposure lost to the ligand (c,d)

Figure 2

Probability distribution of overlap area for native like structures (gray, curve closer to the y-axis) and misdocked structures (black) based on ZDOCK sampling on benchmark 2.0

Figure 3

Size distribution of complexes in the training set (points represented by pluses are bound cases and points represented by circles are unbound cases)

Figure 4

Enzyme Inhibitor – In each row, we present the result of scoring models filtered based on overlap area, the left column is the result of present potential (PIE540), the result of Lu et al (LLS) in the center and that of Tobi et al (TB) is on the right

Figure 5

The same as Figure 4 but this time for Antibody Antigen

Figure 6

The same as Figure 4 but this time for complexes in the Others category

Figure 7

Contribution of Residue contacts (darker shades indicate unfavorable contribution to binding)

Figure 8

Breakdown of contact score by residue type and the contribution of overlap area (OA). The value plotted is the score of the residue type in the native structure averaged over all the 640 complexes in the training set.

Figure 9

Comparison of overlap area and change in surface area computed by DSSP (points represented by pluses are bound cases, points represented by circles are unbound cases and points represented by squares are cases in the benchmark2). Bound cases have almost 0 overlap areas, cases in benchmark2 have low overlap area while unbound cases have higher overlap areas.

Figure 10

Comparison of the residue contact weights for the potentials learnt on the combined set and the set of heterodimers. The value plotted is the score of a contact type in the combined potential vs the score in the potential learnt on the set of heterodimers. The correlation coefficient between the potentials is 0.96 (linear fit for heterodimer potential based on the combined potential has a non zero y-intercept).

Figure 11

Comparison of the residue contact weights for the potentials learnt on the combined set and the set of transient complexes. The value plotted is the score of a contact type in the combined potential vs the score in the potential learnt on the set of transient complexes. The correlation coefficient between the potentials is 0.98