Fig. 3 Structures of designed bis-meP derivatives ( n=2-14 2-14 n=2-14 corrected manually. Hydrogen atoms and Gasteiger-Marsili GOLd docking [35] atomic charges were then added. The structures were minimized in SYBYL for 100 steps with the Tripos force The default parameter settings for three times speed-up field and saved to a multi-MOL2 file were used, and no flipping was allowed In docking runs for pose reproduction, all ten solutions of each ligand were saved for further analysis. However, for virtual screening, Preparation of focused library for virtual screening only the best solution was kept for each molecule. Bis-MEP derivatives with spacer length of two to 14 arbons. a total of 91 molecules were drawn in ISIS/Base Results and discussion with ISIS/Draw from MDL [36] and exported into a 2 D structure data file(SDF)to form a small focused library. Comparison of prediction accuracy of Flexx The 2 D structures of the library were subsequently con- and GOLD verted into 3 D-structures with CORINA [37, still in SDF format. The 3 D structures were then read into SYBYL The accurate prediction of protein-ligand interaction geo- olecular SpreadSheet table for further treatment, such as metries is essential for the success of virtual-screening energy minimization for 100 steps with Tripos force field approaches in structure-based drug design. It requires and Gasteiger-Marsili [35 charges for each molecule. docking tools that are able to generate suitable conforma These structures were put into databases and then written to tions of a ligand within a protein binding site and reliable MOL2 files as input energetic evaluation indicating the quality of the interaction. Flexx [17] uses an incremental construction algorithm where ligands are docked starting with a base fragment Flexx docking The complete ligand is subsequently constructed by adding e remainin omponents. The best solution is selected All Flexx runs used the version 1. ll.I embedded within according to the docking score, using a pure empirical SYBYL. Formal charges were assigned and the Flexx scoring function by Bohm or a knowledge-based scorin scoring function was chosen to evaluate the docking poses. function Drug Score. The GolD [19]( Genetic Optimiza- Table 1. a Fitness scores, RMSD from crystal structures and ICso of known ligands by gold IEVE IVOT IACJ ICso(nM) MOL NAME Fitness RMSD(A) Fitness RMSD(A) Fitness RMSD (A) E2020 66.15 0.5437 48.64 3.3821 46.45 3.7311 13.6 GNT 5753 0.7819 4.8748 4.2323 19954 Half-open BW284c51 80.28 3.5743 85.36 4.1937 63423.6457 0.0036 EDR 44.09 4.2878 43.93 46.02 4.1636 46.35 3.6063 1.21 51550.528 58.4 HUPX 5994 5.3004 59.79 5.0281 0.9428 0.026(K1) THA 45.13 5.8934 49.265.638 50.47 5.6056 From Reference [47] bFrom Reference [48] From Reference [49] dRank 2 scored 50.19 with an rMsD of 0.5404corrected manually. Hydrogen atoms and Gasteiger-Marsili [35] atomic charges were then added. The structures were minimized in SYBYL for 100 steps with the Tripos force field and saved to a multi-MOL2 file. Preparation of focused library for virtual screening Bis-MEP derivatives with spacer length of two to 14 carbons, a total of 91 molecules, were drawn in ISIS/Base with ISIS/Draw from MDL [36] and exported into a 2 D structure data file (SDF) to form a small focused library. The 2 D structures of the library were subsequently con￾verted into 3 D-structures with CORINA [37], still in SDF format. The 3 D structures were then read into SYBYL Molecular SpreadSheet table for further treatment, such as energy minimization for 100 steps with Tripos force field and Gasteiger-Marsili [35] charges for each molecule. These structures were put into databases and then written to MOL2 files as input. FlexX docking All FlexX runs used the version 1.11.1 embedded within SYBYL. Formal charges were assigned and the FlexX scoring function was chosen to evaluate the docking poses. GOLD docking The default parameter settings for three times speed-up were used, and no flipping was allowed. In docking runs for pose reproduction, all ten solutions of each ligand were saved for further analysis. However, for virtual screening, only the best solution was kept for each molecule. Results and discussion Comparison of prediction accuracy of FlexX and GOLD The accurate prediction of protein-ligand interaction geo￾metries is essential for the success of virtual-screening approaches in structure-based drug design. It requires docking tools that are able to generate suitable conforma￾tions of a ligand within a protein binding site and reliable energetic evaluation indicating the quality of the interaction. FlexX [17] uses an incremental construction algorithm where ligands are docked starting with a base fragment. The complete ligand is subsequently constructed by adding the remaining components. The best solution is selected according to the docking score, using a pure empirical scoring function by Böhm or a knowledge-based scoring function DrugScore. The GOLD [19] (Genetic Optimiza￾OH N * OH N * H2 C n (3',3'') n (S,S) n=2~14 (R,S) n=2~14 (R,R) n=2~14 OH N * OH N * * * H N H N H2 C n (3'4',4''3'') n (SR,RS) n=2~14 (SS,SS) n=2~14 (RR,RR) n=2~14 (RS,SR) n=2~14 3' 3'' 3' 4' 3'' 4'' · · · · Fig. 3 Structures of designed bis-MEP derivatives Table 1.a Fitness scores, RMSD from crystal structures and IC50 of known ligands by GOLD 1EVE 1VOT 1ACJ IC50 (nM) MOL_NAME Fitness RMSD (Å) Fitness RMSD (Å) Fitness RMSD (Å) Open E2020 66.15 0.5437 48.64 3.3821 46.45 3.7311 13.6a GNT 57.53 0.7819 54.82 4.8748 53.29 4.2323 1995a Half- open BW284c51 80.28 3.5743 85.36 4.1937 63.42 3.6457 0.0036b EDR 44.09 4.2878 43.93 1.2034 46.02 4.1636 240b HUPA 46.35 3.6063 50.78 1.2146 51.55 0.528 58.4a Closed HUPX 59.94 5.3004 59.79 5.0281 70.48 0.9428 0.026 (Ki)c THA 45.13 5.8934 49.26 5.638 50.47 5.6056d 93a a From Reference [47], b From Reference [48], c From Reference [49], d Rank 2 scored 50.19 with an RMSD of 0.5404 392