Binding affinity prediction with property-encoded shape distribution signatures

Abstract

We report the use of the molecular signatures known as "property-encoded shape distributions" (PESD) together with standard support vector machine (SVM) techniques to produce validated models that can predict the binding affinity of a large number of protein ligand complexes. This "PESD-SVM" method uses PESD signatures that encode molecular shapes and property distributions on protein and ligand surfaces as features to build SVM models that require no subjective feature selection. A simple protocol was employed for tuning the SVM models during their development, and the results were compared to SFCscore, a regression-based method that was previously shown to perform better than 14 other scoring functions. Although the PESD-SVM method is based on only two surface property maps, the overall results were comparable. For most complexes with a dominant enthalpic contribution to binding (DeltaH/-TDeltaS > 3), a good correlation between true and predicted affinities was observed. Entropy and solvent were not considered in the present approach, and further improvement in accuracy would require accounting for these components rigorously.

Figure 1

(Left) Side view of EP mapped protein interaction surface of complex 1fbp. P1 and P2 are two points chosen from random locations on the surface. The properties of these two points and the Euclidean distance d between them determine which PESD signature bin they will occupy. The graphical representation of the PESD signature of 1fbp is shown as a two dimensional grid of bins (Right). Darker circles indicate greater bin populations. Each row corresponds to specific endpoint color combinations while each column represents point-pair distances that increase from left to right.

Figure 2

H_1-4 depict protein and ligand interaction surfaces encoded with EP and Active LP maps. PESD signatures derived from these surfaces were used as features for building binding affinity SVM models.

Figure 3

Plot of experimental affinities versus predicted affinities for PESD-SVM regression models I, II and V applied to their respective test sets.

Figure 4

Plot of R_P, R_S and number of test cases against chi-squared cutoff distances for PESD-SVM regression Model II.

Figure 5

Correlation between ligand pose (rmsd from native pose) and PESD-DOCK SVM Score for docked conformations of L-Benzyl succinate in 1cbx. The score of the native pose (rmsd = 0) is shown as a dashed line. Since PESD-DOCK SVM models were trained on positive affinity values (pK_d/pK_i), higher scores indicate favorable interactions.

Figure 6

(a) Plot of free energy versus enthalpy for 322 entries from the SCORPIO database. (b) Plot of free energy versus enthalpy for 111 out of 322 entries from the SCORPIO database. The ΔH/-TΔS was greater than 3 for these complexes. (c) Difference plot of 6a and 6b (d) Plot of experimental versus predicted affinities of all entries in the core set for which ΔH/-TΔS was greater than 3 and could be obtained from the SCORPIO database. The complex 1adl is circled in red.