The Compressed Feature Matrix—a novel descriptor for adaptive similarity search

Abstract

The Compressed Feature Matrix (CFM) is a new molecular descriptor for adaptive similarity searching. Depending on the requirements, it is based on a distance or geometry matrix. Thus, the CFM permits topological and three-dimensional comparisons of molecules. In contrast to the common distance matrix, the CFM is based on features instead of atoms. Each kind of these features may be weighted separately, depending on its (estimated) contribution to the biological effect of the molecule. In this work, we show that the CFM allows us to adapt similarity evaluations to particular ligands as well as to classification requirements. The CFM method is analyzed regarding correctness, adaptivity and speed. Applying the basic setting of feature weights, the similarity evaluations using the CFM on the one hand and the Tanimoto coefficient together with MACCS Keys on the other yield similar results. However, in contrast to the latter method, the CFM even permits us to focus on small parts of molecules to serve as a basis for similarity. Accordingly, we have achieved striking results not only by readjusting the feature weights with regard to the scaffold but also to the side chain of the respective target. The results of the latter run turned out to be rather independent of the molecular scaffold. Hence, the CFM is suitable not only for common similarity evaluation, but also for techniques such as lead or scaffold hopping. Figure Chemical structure, feature graph and topological CFM of serotonine

Kernel Functions for Attributed Molecular Graphs – A New Similarity‐Based Approach to ADME Prediction in Classification and Regression

“…In section 4 we experimentally evaluate our approach and compare it to classical descriptor-based QSAR models. Our experiments include prediction of HIA [25 -31], Blood-Brain-Barrier (BBB) penetration [32,33], bioavailability [34], and grouping inhibitors into four different classes [35]. We show that by using our approach we achieve a generalization performance comparable to a descriptor-based model, which includes only descriptors that are a-priori known to be relevant for the problem.…”

“…Finally, we investigated a set of 296 molecules published in [35] as a test dataset for the SOL project 4 . The dataset consists of 4 different classes of inhibitors: thrombin inhibitors (75 molecules), serotonin inhibitors of the 5HT2 class (75 molecules), monoamine oxidase inhibitors (71 molecules) and 5-hydroxytryptamine oxidase (75 molecules).…”

A 3D Similarity Method for Scaffold Hopping from Known Drugs or Natural Ligands to New Chemotypes

“…1,[3][4][5][6][7][8][9][10][11][12][13] The ability to move to new scaffolds can be of interest in situations where the natural ligands or substrates of protein targets are known but synthetic inhibitors are not and structural information about the target protein is not available. An ideal similarity search method could use endogenous ligand structures to discover drug-like mimetics in large databases in an automated manner.…”

Information Is Selection—A Review of Basics Shows Substantial Potential for Improvement of Digital Information Representation

“…Therefore, it is important that the users-especially experts in a certain topic-can define terms in this topic in the adapted domain with the topic-specific relevant features that they want as search criteria. The use of relevant features as criteria for similarity searches has been, up to now, a typical application in databases [28][29][30][31][32][33][34][35][36][37][38]. This restriction, however, is not necessary.…”