MDDR database

We have evaluated the various approaches described above by means of simulated virtual screening searches on the MDL Drug Data Report (MDDR) database. After removal of duplicates and molecules that could not be processed using local software, a total of 102 535 molecules were available for searching. These molecules were represented by 988-bit Tripos Unity 2D fingerprints, and searched using the eleven sets of active compounds detailed in Table 1. 

We thank the following Novartis Institutes for BioMedical Research for funding JH MDL Information Systems Inc. for the provision of the MDDR database Barnard Chemical Information Ltd., Daylight Chemical Information Systems Inc., the Royal Society, Tripos Inc. and the Wolfson Foundation for software and laboratory support. 

Both enrichment curves show a steep beginning, almost parallel to the ideal curve (black line). The flattening of the curves towards the right can be explained by the fact that some a1A compounds of the MDDR database cannot be mapped by the pharmacophore and thus obtain fit values of 0. The steepness of the enrichment curve on the left, however, reflects, that among the top-ranked... 

The 3-point and 4-point pharmacophore methods can be used to analyse and compare different sets of compounds and databases. Figure 7 illustrates the 4-point pharmacophores for the MDDR database [25], the Available Chemical Directory (ACD) [25], a company registry database and a set of combinatorial libraries reported by Mason and co-workers [7, 11, 13]. Previous studies [3] had shown the increase in resolution possible using 4-point instead of 3-point pharmacophores. 

The example reported by Pearlman and co-workers [18, 19, 26] involves the analysis of ACE inhibitors. They found that 3 of the 6 BCUT metrics they had identified from an analysis of the MDDR database (60,000 drug compounds) were receptor relevant , i.e. that the actives clustered in these dimensions. Figure 16 shows the actives clustering in a sea of general drugs (MDRR compounds) in these 3 dimensions. 

Figure 5.7. Comparisons of the 3D four-point pharmacophore fingerprints exhibited by several sets [MDDR database of 62,000 biologically active compounds, a corporate registry database of 100,000 compormds used for screening, 100,000 compounds from combinatorial libraries (from a four-com-poilent Ugi condensation reaction), and 14,000 compound random subsets (MDDR, corporate) or indlividual libraries]. The four-point potential pharmacophores were calculated using 10 distance range bins and the standard six pharmacophore features.

Figure 5.21. Examples of 7-TM GPCR leged" motifs found in the MDDR database.

The VS studies of FXa inhibitors were firstly conducted against the annotated MDDR database by using the developed LEVS models. The VS performance of the developed SVM and RF models is given in Table 8.3. As shown in the table, the developed SVM and RF models can identify the substantially high percentages of 40.59 % and 72.99 % of the known FXa inhibitors from the MDDR database, respectively, in which there are in total 1318 known FXa inhibitors. The hit rates of the developed SVM and RF models are 3.42 % and 3.89 %, respectively, which are significantly improved against the random hit rate of 0.87 % (1318/151753 = 0.0087). 

Table 8.3 The VS performance of S VM and RF for FXa inhibitors against the MDDR database and the fragment-like subset of ZINC database...

Method The MDDR database The fragment-Kke subset of ZINC database ... 

The enrichment factors of the developed SVM and RF models are 3.93 and 4.48, respectively, which are comparable to those in the reported study [55], in which the averaged enrichment factor of around 4 was obtained over several classes of active compounds derived from the MDDR database. The relatively low hit rates and enrichment factors probably attribute to a large number of false positives. Therefore, it is desirable to further reduce the number of false positives by the combination of other LEVS and SBVS approaches. 

Finally, the M LCC score of a compound is defined as the highest level n at which the compound is found compatible with the reference drug library, or zero if no compatibility is found at any level. This method was tested by using 11704 drugs from CMC and MDDR databases. 

M.J.K is supported by a National Science Foundation graduate fellowship. J.H. is supported by the sixth Framework Program of the European Commission. We are grateful to MDL Information Systems Inc. for the MDDR database Daylight Chemical Information Systems Inc. and OpenEye Scientific Software for software support. We thank John J. Irwin for reading the manuscript and Brian K. Shoichet for mentoring. 

Fig. 15.4-6 Analysis of privileged scaffold-target matrix of monoamine CPCR ligands. For each CPCR ligand assigned in the MDL Drug Data Report (MDDR) database to a specific monoamine CPCR subtype, the Bemis-Murcko frameworks were generated. The lists of frameworks were then combined and duplicates were eliminated. The comprehensive list of unique frameworks define the row vector of the matrix, and the CPCR subtypes were arranged to the column vector. The matrix...

Several types of artificial neural networks have been used as pattern recognition engines. In general, neural networks are mathematical models that emulate some of the observed properties of biological nervous systems, in particular the ability to learn complex relationships. Examples of drug-like molecules were taken from the WDI, CMC or MDDR databases, while compounds from the ACD database exemplified nondrugs. Molecules were represented by a large number of descriptors, such as ISIS... 

---