Ever since the first automated de novo design techniques were conceived only 15 years ago, the computer-based design of hit and lead structure candidates has emerged as a complementary approach to high-throughput screening. Although many challenges remain, de novo design supports drug discovery projects by generating novel pharmaceutically active agents with desired properties in a cost- and time-efficient manner. In this review, we outline the various design concepts and highlight current developments in computer-based de novo design.

No abstract

Support vector machine (SVM) and artificial neural network (ANN) systems were applied to a drug/nondrug classification problem as an example of binary decision problems in early-phase virtual compound filtering and screening. The results indicate that solutions obtained by SVM training seem to be more robust with a smaller standard error compared to ANN training. Generally, the SVM classifier yielded slightly higher prediction accuracy than ANN, irrespective of the type of descriptors used for molecule encoding, the size of the training data sets, and the algorithm employed for neural network training. The performance was compared using various different descriptor sets and descriptor combinations based on the 120 standard Ghose-Crippen fragment descriptors, a wide range of 180 different properties and physicochemical descriptors from the Molecular Operating Environment (MOE) package, and 225 topological pharmacophore (CATS) descriptors. For the complete set of 525 descriptors cross-validated classification by SVM yielded 82% correct predictions (Matthews cc = 0.63), whereas ANN reached 80% correct predictions (Matthews cc = 0.58). Although SVM outperformed the ANN classifiers with regard to overall prediction accuracy, both methods were shown to complement each other, as the sets of true positives, false positives (overprediction), true negatives, and false negatives (underprediction) produced by the two classifiers were not identical. The theory of SVM and ANN training is briefly reviewed.

Correlation vector methods were tested for their usefulness in ligand-based virtual screening. Three molecular descriptors--two based on potential pharmacophore points and one on partial atom charges--and three similarity measures--the Manhattan distance, the Euclidian distance and the Tanimoto coefficient--were compared. The alignment-free descriptors seem to be particularly applicable when a course-grain filtering of data sets is required in combination with a high execution speed. Significant enrichment of actives was obtained by retrospective analysis. The cumulative percentages for all three descriptors allow for the retrieval of up to 78% of the active molecules in the first five percent of the reference database. Different descriptors retrieved only weakly overlapping sets of active molecules among the top-ranking compounds. If a single similarity index is to be used, the Manhattan distance seems to be particularly applicable. Generally, none of the three different descriptors tested in this study clearly outperformed the others. The suitability of a descriptor critically depends on the ligand-receptor interaction under investigation. For ligand-based similarity searching it is recommended to exploit several descriptors in parallel.