Parameter Estimation for Scoring Protein−Ligand Interactions Using Negative Training Data

Abstract

Surflex-Dock employs an empirically derived scoring function to rank putative protein-ligand interactions by flexible docking of small molecules to proteins of known structure. The scoring function employed by Surflex was developed purely on the basis of positive data, comprising noncovalent protein-ligand complexes with known binding affinities. Consequently, scoring function terms for improper interactions received little weight in parameter estimation, and an ad hoc scheme for avoiding protein-ligand interpenetration was adopted. We present a generalized method for incorporating synthetically generated negative training data, which allows for rigorous estimation of all scoring function parameters. Geometric docking accuracy remained excellent under the new parametrization. In addition, a test of screening utility covering a diverse set of 29 proteins and corresponding ligand sets showed improved performance. Maximal enrichment of true ligands over nonligands exceeded 20-fold in over 80% of cases, with enrichment of greater than 100-fold in over 50% of cases.

Surflex-Dock 2.1: Robust performance from ligand energetic modeling, ring flexibility, and knowledge-based search

“…Pham and Jain [17] reported performance of Surflex-Dock Version 1.3 on 29 proteins, each with known ligands ranging in number from 5 to 20. Two decoy sets were used, one being a modified version of that made available by Rognan's group [18,19], and the other being derived from the ZINC database.…”

“…Implicit in this treatment is the idea that the space covered by varying the pose parameters alone includes a high-scoring ligand pose that is close to correct. Earlier versions of Surflex-Dock were restricted purely to exploration of this pose parameter space [16,17]. The focus of this work is in moving beyond the limitations of exploring the pure pose parameter search space and, when appropriate, in exploring it with the benefit of prior knowledge of a protein ligand interaction.…”

“…The latter measures the ability of a docking algorithm, given a single protein structure, to correctly rank a set of known active ligands against a background of putative inactives (called decoys). A recent and valuable trend within the field has been the use of standard benchmarks on multiple methods [16][17][18][19][20][21][22]. For the work reported here, benchmarks were selected either for which previous versions of Surflex-Dock had been tested [16,17,19] in order to show the effects of new features, or where protein and ligand structures were publicly available along with performance of widely used methods [23,24].…”

“…Screening utility, 27 proteins: This set consists of 27 protein structures, each with from 5 to 20 known ligands along with two different sets of decoy molecules. The set incorporates the widely used estrogen receptor and thymidine kinase test cases of Rognan's group [13,[16][17][18][19]. This will be called the Pham set.…”

Impact of Benchmark Data Set Topology on the Validation of Virtual Screening Methods: Exploration and Quantification by Spatial Statistics

“…13,14 In order to avoid these issues, the area under the receiver operating characteristic curve (ROC) was used in a number of studies for the analysis of VS validation rankings. [15][16][17][18] However, the ROC metric has one important shortcoming: it is unable to address the so-called "early recognition" problem. 13,14 Since usually only a small fraction of a VS ranking can be tested experimentally, a good metric for VS should reflect the enrichment of actives at the beginning of the ranking.…”