Computer-aided prediction of receptor profile for drug-like compounds

Computer program SIMEST (Similarity Estimating) for prompt estimation of drug-like compounds' interaction with various receptors has been developed. More than 200 kinds of receptor activities (agonists and antagonists) can be predicted analysing new compound's similarity with known ligands. SIMEST consists of two components: (a) database of low-molecular weight ligands (agonists and antagonists) for 106 receptors and (b) similarity estimation module based on original MNA descriptors. The predictive abilities of SIMEST were evaluated on two data sets: (a) about 24,000 biologically active compounds from MDDR-99.2 database; (b) seven well-studied antipsychotic drugs. Average accuracy of prediction was 85.2% for compounds from MDDR and 78.7% for antipsychotic drugs. SIMEST can be effectively applied for predicting possible mechanisms of action and side effects for drug-like substances on the basis of their structural formulae.