Contribution of 2D and 3D Structural Features of Drug Molecules in the Prediction of Drug Profile Matching

Abstract

Drug Profile Matching (DPM), a novel virtual affinity fingerprinting method capable of predicting the medical effect profiles of small molecules, was introduced by our group recently. The method exploits the information content of interaction patterns generated by flexible docking to a series of rigidly kept nontarget protein active sites. We presented the ability of DPM to classify molecules excellently, and the question arose, what the contribution of 2D and 3D structural features of the small molecules is to the intriguingly high prediction power of DPM. The present study compared the prediction powers for effect profiles of 1163 FDA-approved drug compounds determined by DPM and ChemAxon 2D and 3D similarity fingerprinting approaches. We found that DPM outperformed the 2D and 3D approaches in almost all therapeutic categories for drug classification except for mechanically rigid structural categories where high accuracy was obtained by all three methods. Moreover, we also tested the predictive power of DPM on external data by reducing the parent data set and demonstrated that DPM can overcome the common screening problems of 2D and 3D similarity methods arising from the presence of structurally diverse molecules in certain effect categories.

Identification of PPARγ ligands with One-dimensional Drug Profile Matching

“…In a following study, we studied the contribution of the structural features of the drugs to the high prediction power of the method. We showed that DPM outperformed conventional two-dimensional and three-dimensional structural similarity-based prediction approaches in almost all examined categories 4. Nevertheless, its prediction accuracy is limited if only a few compounds can be used as a learning set for a given effect.…”

Progress in the Analysis of Multiple Activity Profile of Screening Data Using Computational Approaches

“…[ Peragovics et al, 2012;Simon et al, 2012] Drug-target interactions from chemical and genomic kernels using Bayesian matrix factorization Four drug-target interaction networks involving enzymes, ion channels, G-protein-coupled receptors and nuclear receptors were considered. Authors proposed a novel Bayesian formulation that combines dimensionality reduction, matrix factorization and binary classification for predicting drug-target interaction networks using only chemical similarity between drug compounds and genomic similarity between target proteins.…”

Prediction of Protein−compound Binding Energies from Known Activity Data: Docking‐score‐based Method and its Applications

“…The quantitative structure−activity relationship (QSAR) approach is a useful tool for optimizing leads and predicting target/off‐target activities and toxicity. QSAR‐based affinity predictions are useful for the general drug development process, including the repositioning (repurposing) of already approved drugs, poly‐pharmacology, and the prediction of drug−drug interactions . The recent accumulation of protein−compound affinity data in public repositories, such as the PubChem and ChEMBL projects, has enabled us to carry out proteome‐wide target/off‐target predictions ,.…”

“…QSAR-based affinity predictions are useful for the general drug development process, including the repositioning (repurposing) of already approved drugs, poly-pharmacology, and the prediction of drugÀdrug interactions. [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15] The recent accumulation of proteinÀcompound affinity data in public repositories, such as the PubChem and ChEMBL projects, has enabled us to carry out proteome-wide target/off-target predictions. [16,17] These predictions are based on QSAR models for multiple proteins, just as in conventional computer-aided drug design and virtual screening.…”

“…Wide application of QSAR-based models in computeraided drug development, such as proteinÀcompound binding free energy (affinity) prediction, target/off-target predictions, and counter screening based on QSAR models, has succeeded in many studies, including ours. [3,4,7,8] Most QSAR models rely on descriptors with sets of two-dimensional (2D) substructures; the most popular such descriptors are MDL's MACCS key and 0-3D molecular descriptors (e.g., 5,270 descriptors recorded in Dragon (Kode srl, Pisa, Italy)). In our previous studies, we developed QSAR methods for the affinity prediction of a compound by using docking studies against multiple proteins.…”