A Computational Method for Unveiling the Target Promiscuity of Pharmacologically Active Compounds

Abstract

Drug discovery is governed by the desire to find ligands with defined modes of action. It has been realized that even designated selective drugs may have more macromolecular targets than is commonly thought. Consequently, it will be mandatory to consider multitarget activity for the design of future medicines. Computational models assist medicinal chemists in this effort by helping to eliminate unsuitable lead structures and spot undesired drug effects early in the discovery process. Here, we present a straightforward computational method to find previously unknown targets of pharmacologically active compounds. Validation experiments revealed hitherto unknown targets of the natural product resveratrol and the nonsteroidal anti-inflammatory drug celecoxib. The obtained results advocate machine learning for polypharmacology-based molecular design, drug re-purposing, and the "de-orphaning" of phenotypic drug effects.

Virtual Screening and Design with Machine Intelligence Applied to Pim‐1 Kinase Inhibitors

“…Software. Software for molecular descriptor calculation (CATS), [8] self-organizing map training and analysis (POOMA), [19] target prediction (TIGER), [20] and similarity searching (inSili.com LLC, Zurich, Switzerland) was run on an iMac workstation. Parameter settings: CATS: feature correlation over 0-9 bonds, with type scaling; POOMA: 20 × 10 toroidal map, Gaussian neighborhood, random initialization, linear decay of the learning rate (τ initial = 1), 10 9 learning cycles.…”

Polypharmacological Drug−target Inference for Chemogenomics

“…Briefly, the algorithm considers all compounds of the winning SOM cluster (Voronoi field) as relevant for similarity assessment, and computes two scores for the query compound. The log( Odds ) score estimates the information content of a taget prediction, and the AUC score considers the confidence of a prediction by weighting the drugs from the immediate vicinity of the query higher than the more distant drugs within the SOM cluster . This scoring process is performed for each of the two SOMs, and the final TIGER score is calculated as the averaged product of the individual scores.…”

The Polypharmacology Browser PPB2: Target Prediction Combining Nearest Neighbors with Machine Learning

“…LBVS also forms the basis for target prediction tools, which assign possible targets to a molecule based on its similarity to known, target annotated molecules such as those in the ChEMBL database. 2 Targets are assigned either directly based on nearest neighbor (NN) relationships, or indirectly by building a machine learning (ML) model, [3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22] with several tools available online. [23][24][25][26][27][28][29][30][31][32][33] Herein we report PPB2 (Polypharmacology Browser version 2) as an extension of our previously reported web-portal PPB (Polypharmacology Browser).…”