Limits...
Pharmacophore-based similarity scoring for DOCK.

Jiang L, Rizzo RC - J Phys Chem B (2014)

Bottom Line: When used alone, FMS scoring dramatically improves pose reproduction success to 93.5% (∼20% increase) and reduces sampling failures to 3.7% (∼6% drop) compared to the standard energy score (SGE) across 1043 protein-ligand complexes.The combined FMS+SGE function further improves success to 98.3%.For enrichment, incorporating pharmacophores during sampling and scoring, in most cases, also yield improved outcomes when docking and rank-ordering libraries of known actives and decoys to 15 systems.

View Article: PubMed Central - PubMed

Affiliation: Department of Applied Mathematics & Statistics, ‡Institute of Chemical Biology & Drug Discovery, §Laufer Center for Physical & Quantitative Biology, Stony Brook University , Stony Brook, New York 11794-3600, United States.

ABSTRACT
Pharmacophore modeling incorporates geometric and chemical features of known inhibitors and/or targeted binding sites to rationally identify and design new drug leads. In this study, we have encoded a three-dimensional pharmacophore matching similarity (FMS) scoring function into the structure-based design program DOCK. Validation and characterization of the method are presented through pose reproduction, crossdocking, and enrichment studies. When used alone, FMS scoring dramatically improves pose reproduction success to 93.5% (∼20% increase) and reduces sampling failures to 3.7% (∼6% drop) compared to the standard energy score (SGE) across 1043 protein-ligand complexes. The combined FMS+SGE function further improves success to 98.3%. Crossdocking experiments using FMS and FMS+SGE scoring, for six diverse protein families, similarly showed improvements in success, provided proper pharmacophore references are employed. For enrichment, incorporating pharmacophores during sampling and scoring, in most cases, also yield improved outcomes when docking and rank-ordering libraries of known actives and decoys to 15 systems. Retrospective analyses of virtual screenings to three clinical drug targets (EGFR, IGF-1R, and HIVgp41) using X-ray structures of known inhibitors as pharmacophore references are also reported, including a customized FMS scoring protocol to bias on selected regions in the reference. Overall, the results and fundamental insights gained from this study should benefit the docking community in general, particularly researchers using the new FMS method to guide computational drug discovery with DOCK.

Show MeSH
Validation metrics used to evaluate DOCK scoring functions. (a)Pose reproduction cases with different outcomes: Success (top, PDBcode 3CPA),Score Fail (middle, PDB code 1V2W), and Sample Fail (bottom, PDB code 1GKC). Crystal posesin orange, best scored poses in magenta, best RMSD pose in cyan. (b)Representative crossdocking heatmap showing docking outcome as a functionof docking all ligands (Lig1, Lig2, ..., LigN) to all receptors (Rec1,Rec2, ..., RecN) for an aligned group of proteins with nearly identicalsequence homology. (c) Hypothetical database enrichment results showinga partitioning of data based on FMS score ranking (0 to 6) for a groupof ligands (left bottom, magenta curve) comprised of a known activeligand set (left middle, blue curve) and inactive decoy set (lefttop, red curve). The vertical dashed line represents a hypotheticalFMS score cutoff dividing the total group into (X) predicted positive and (Y) predicted negativesets which can be partitioned into four quadrants (I–IV) definedrespectively as true positives (TP, I), false positives (FP, II),true negatives (TN, III), and false negatives (FN, IV). Also shownis an ROC curve, which for this example plots individual points whichcorrespond to various FMS score cutoffs in the left panel. The coordinateof each point is determined by the false positive rate and true positiverate at that FMS score cutoff.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4306494&req=5

fig6: Validation metrics used to evaluate DOCK scoring functions. (a)Pose reproduction cases with different outcomes: Success (top, PDBcode 3CPA),Score Fail (middle, PDB code 1V2W), and Sample Fail (bottom, PDB code 1GKC). Crystal posesin orange, best scored poses in magenta, best RMSD pose in cyan. (b)Representative crossdocking heatmap showing docking outcome as a functionof docking all ligands (Lig1, Lig2, ..., LigN) to all receptors (Rec1,Rec2, ..., RecN) for an aligned group of proteins with nearly identicalsequence homology. (c) Hypothetical database enrichment results showinga partitioning of data based on FMS score ranking (0 to 6) for a groupof ligands (left bottom, magenta curve) comprised of a known activeligand set (left middle, blue curve) and inactive decoy set (lefttop, red curve). The vertical dashed line represents a hypotheticalFMS score cutoff dividing the total group into (X) predicted positive and (Y) predicted negativesets which can be partitioned into four quadrants (I–IV) definedrespectively as true positives (TP, I), false positives (FP, II),true negatives (TN, III), and false negatives (FN, IV). Also shownis an ROC curve, which for this example plots individual points whichcorrespond to various FMS score cutoffs in the left panel. The coordinateof each point is determined by the false positive rate and true positiverate at that FMS score cutoff.

Mentions: Representative visual examples of the three outcomesare shown in Figure 6a. For ligands of druglikesize, low RMSD values also typically correspond to good visual overlapbetween docked and reference ligand poses. All statistics reportedin this work make use of “symmetry corrected” RMSDsto account for chemically identical functionality (i.e., symmetricring flips, carboxylate flips, etc.) or completely symmetric molecules,adopting visually indistinguishable conformations as described indetail previously.37 The updated pose reproductiondatabase termed SB2012 (an update of the SB2010 database)27 was used for all pose reproduction and crossdocking(defined below) experiments. The set, derived from complexes in theprotein databank (PDB), contains 1043 protein–ligand systemsin ready to DOCK format and is freely available online at www.rizzolab.org.


Pharmacophore-based similarity scoring for DOCK.

Jiang L, Rizzo RC - J Phys Chem B (2014)

Validation metrics used to evaluate DOCK scoring functions. (a)Pose reproduction cases with different outcomes: Success (top, PDBcode 3CPA),Score Fail (middle, PDB code 1V2W), and Sample Fail (bottom, PDB code 1GKC). Crystal posesin orange, best scored poses in magenta, best RMSD pose in cyan. (b)Representative crossdocking heatmap showing docking outcome as a functionof docking all ligands (Lig1, Lig2, ..., LigN) to all receptors (Rec1,Rec2, ..., RecN) for an aligned group of proteins with nearly identicalsequence homology. (c) Hypothetical database enrichment results showinga partitioning of data based on FMS score ranking (0 to 6) for a groupof ligands (left bottom, magenta curve) comprised of a known activeligand set (left middle, blue curve) and inactive decoy set (lefttop, red curve). The vertical dashed line represents a hypotheticalFMS score cutoff dividing the total group into (X) predicted positive and (Y) predicted negativesets which can be partitioned into four quadrants (I–IV) definedrespectively as true positives (TP, I), false positives (FP, II),true negatives (TN, III), and false negatives (FN, IV). Also shownis an ROC curve, which for this example plots individual points whichcorrespond to various FMS score cutoffs in the left panel. The coordinateof each point is determined by the false positive rate and true positiverate at that FMS score cutoff.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4306494&req=5

fig6: Validation metrics used to evaluate DOCK scoring functions. (a)Pose reproduction cases with different outcomes: Success (top, PDBcode 3CPA),Score Fail (middle, PDB code 1V2W), and Sample Fail (bottom, PDB code 1GKC). Crystal posesin orange, best scored poses in magenta, best RMSD pose in cyan. (b)Representative crossdocking heatmap showing docking outcome as a functionof docking all ligands (Lig1, Lig2, ..., LigN) to all receptors (Rec1,Rec2, ..., RecN) for an aligned group of proteins with nearly identicalsequence homology. (c) Hypothetical database enrichment results showinga partitioning of data based on FMS score ranking (0 to 6) for a groupof ligands (left bottom, magenta curve) comprised of a known activeligand set (left middle, blue curve) and inactive decoy set (lefttop, red curve). The vertical dashed line represents a hypotheticalFMS score cutoff dividing the total group into (X) predicted positive and (Y) predicted negativesets which can be partitioned into four quadrants (I–IV) definedrespectively as true positives (TP, I), false positives (FP, II),true negatives (TN, III), and false negatives (FN, IV). Also shownis an ROC curve, which for this example plots individual points whichcorrespond to various FMS score cutoffs in the left panel. The coordinateof each point is determined by the false positive rate and true positiverate at that FMS score cutoff.
Mentions: Representative visual examples of the three outcomesare shown in Figure 6a. For ligands of druglikesize, low RMSD values also typically correspond to good visual overlapbetween docked and reference ligand poses. All statistics reportedin this work make use of “symmetry corrected” RMSDsto account for chemically identical functionality (i.e., symmetricring flips, carboxylate flips, etc.) or completely symmetric molecules,adopting visually indistinguishable conformations as described indetail previously.37 The updated pose reproductiondatabase termed SB2012 (an update of the SB2010 database)27 was used for all pose reproduction and crossdocking(defined below) experiments. The set, derived from complexes in theprotein databank (PDB), contains 1043 protein–ligand systemsin ready to DOCK format and is freely available online at www.rizzolab.org.

Bottom Line: When used alone, FMS scoring dramatically improves pose reproduction success to 93.5% (∼20% increase) and reduces sampling failures to 3.7% (∼6% drop) compared to the standard energy score (SGE) across 1043 protein-ligand complexes.The combined FMS+SGE function further improves success to 98.3%.For enrichment, incorporating pharmacophores during sampling and scoring, in most cases, also yield improved outcomes when docking and rank-ordering libraries of known actives and decoys to 15 systems.

View Article: PubMed Central - PubMed

Affiliation: Department of Applied Mathematics & Statistics, ‡Institute of Chemical Biology & Drug Discovery, §Laufer Center for Physical & Quantitative Biology, Stony Brook University , Stony Brook, New York 11794-3600, United States.

ABSTRACT
Pharmacophore modeling incorporates geometric and chemical features of known inhibitors and/or targeted binding sites to rationally identify and design new drug leads. In this study, we have encoded a three-dimensional pharmacophore matching similarity (FMS) scoring function into the structure-based design program DOCK. Validation and characterization of the method are presented through pose reproduction, crossdocking, and enrichment studies. When used alone, FMS scoring dramatically improves pose reproduction success to 93.5% (∼20% increase) and reduces sampling failures to 3.7% (∼6% drop) compared to the standard energy score (SGE) across 1043 protein-ligand complexes. The combined FMS+SGE function further improves success to 98.3%. Crossdocking experiments using FMS and FMS+SGE scoring, for six diverse protein families, similarly showed improvements in success, provided proper pharmacophore references are employed. For enrichment, incorporating pharmacophores during sampling and scoring, in most cases, also yield improved outcomes when docking and rank-ordering libraries of known actives and decoys to 15 systems. Retrospective analyses of virtual screenings to three clinical drug targets (EGFR, IGF-1R, and HIVgp41) using X-ray structures of known inhibitors as pharmacophore references are also reported, including a customized FMS scoring protocol to bias on selected regions in the reference. Overall, the results and fundamental insights gained from this study should benefit the docking community in general, particularly researchers using the new FMS method to guide computational drug discovery with DOCK.

Show MeSH