Identifying the macromolecular targets of de novo-designed chemical entities through self-organizing map consensus

Abstract

De novo molecular design and in silico prediction of polypharmacological profiles are emerging research topics that will profoundly affect the future of drug discovery and chemical biology. The goal is to identify the macromolecular targets of new chemical agents. Although several computational tools for predicting such targets are publicly available, none of these methods was explicitly designed to predict target engagement by de novo-designed molecules. Here we present the development and practical application of a unique technique, self-organizing map-based prediction of drug equivalence relationships (SPiDER), that merges the concepts of self-organizing maps, consensus scoring, and statistical analysis to successfully identify targets for both known drugs and computer-generated molecular scaffolds. We discovered a potential off-target liability of fenofibraterelated compounds, and in a comprehensive prospective application, we identified a multitarget-modulating profile of de novo designed molecules. These results demonstrate that SPiDER may be used to identify innovative compounds in chemical biology and in the early stages of drug discovery, and help investigate the potential side effects of drugs and their repurposing options.drug design | target prediction | polypharmacology | machine learning | chemical similarity C omputer-assisted de novo molecular design has evolved as a popular source of ideas to combat the perceived lack of new chemical entities (NCEs) in chemical biology and drug discovery (1). We demonstrate that automated de novo design delivers readily synthesizable NCEs with desirable activity profiles. Although receptor-based design operates on a model of a macromolecular binding site, ligand-based methods are either explicitly or implicitly based on the chemical similarity principle (2) without requiring a receptor model (3). Instead, the latter typically uses some measure of chemical or pharmacophore feature similarity to a reference ligand as a fitness function, which aims to generate NCEs as template mimetics via scaffold hopping (4-8). We report the development, implementation, and successful prospective application of an innovative computational technique for the target profiling of de novo-designed molecules. The approach combines the concepts of self-organizing maps (SOMs) (9) for macromolecular target prediction (10), consensus scoring (11), and a statistical evaluation that provides confidence estimates for the predictions.Predicting polypharmacological activities is a topic relevant to chemical biology and drug discovery, not only to take advantage of inherent drug promiscuity but also to decrease lead compound attrition caused by unfavorable off-target modulation

Revealing the macromolecular targets of complex natural products

“…To prioritize the candidate macromolecular targets, we computed confidence scores for each prediction. This ordering concept has been shown to be useful for prospectively identifying targets of new chemical entities (NCEs), which is conceptually difficult because NCEs reside outside the applicability domain of purely structure-based methods 23 . We chose this approach because we expected the target prediction for natural products to be equally challenging 10 .…”

Active Learning for Computational Chemogenomics

“…It would then appear that in certain cases, curiosity selection is building local SAR models for specific targets in spurts. The idea of many per-target quantitative structure-activity relationship models as a chemogenomic model has been explored previously [21,[92][93][94][95][96]. A key difference between these www.future-science.com 399 future science group Active learning for computational chemogenomics Research Article per-target approaches and the approach explored here is that we have removed the requirement to have a sufficient number of ligands per target in the per-target models, under the presumption that a sufficient number of similar ligand-target pairs also have similar bioactivity.…”

“…The choice for either approach is strongly governed by data availability or simply personal preference, with no clear winner among the numerous retrospective comparisons or when reviewing the literature on prospective applications [16][17][18]. The benefit of using complementary approaches has been investigated previously and justifies the existence of a multitude of methods that have distinct applicability domains [19][20][21].…”

Computational Polypharmacology Analysis of the Heat Shock Protein 90 Interactome

“…3,5 To this end, an optimal selection of target combinations to be used in polypharmacology approaches is crucial. Methods aiming at the identification of such compounds have been recently proposed, [11][12][13][14][15] and other approaches are currently being explored. 5 Here, we propose a computational strategy aiming at the identification of promising target combinations for a polypharmacology approach.…”