Unsupervised pharmacophore modeling combined with QSAR analyses revealed novel low micromolar SIRT2 inhibitors

Abstract

Situin 2 (SIRT2) enzyme is a histone deacetylase that has important role in neuronal development. SIRT2 is clinically validated target for neurodegenerative diseases and some cancers. In this study, exhaustive unsupervised pharmacophore modeling was combined with quantitative structure-activity relationship (QSAR) analysis to explore the structural requirements for potent SIRT2 inhibitors using 146 known SIRT2 ligands. A computational workflow that combines genetic function algorithm with k-nearest neighbor or multiple linear regression was implemented to build self-consistent and predictive QSAR models based on combinations of pharmacophores and physicochemical descriptors. Successful pharmacophores were complemented with exclusion spheres to optimize their receiver operating characteristic curve profiles. Optimal QSAR models and their associated pharmacophore hypotheses were experimentally validated by identification and in vitro evaluation of several new promising SIRT2 inhibitory leads retrieved from the National Cancer Institute structural database. The most potent hit illustrated IC value of 5.4μM. The chemical structures of active hits were validated by proton nuclear magnetic resonance and mass spectroscopy.

Discovery of new potent lysine specific histone demythelase-1 inhibitors (LSD-1) using structure based and ligand based molecular modelling and machine learning

“…58 In this context, ligand-based pharmacophore modelling can either proceed guided by the bioactivities of training compounds (i.e., supervised modelling) or guided by commonalities of binding features among training compounds (i.e., unsupervised modelling). 59 In this project, we relied on the HYPOGEN module of Discovery Studio soware suite (version 4.5) for supervised pharmacophore modelling. 60 Moreover, it was decided to use the "Common Feature Pharmacophore Generation" (also known as HIPHOP) protocol within Discovery Studio for unsupervised ligand-based pharmacophore modelling, i.e., to identify common binding features among potent ligands independent of their bioactivities, 59 which allowed us to incorporate additional training compounds irrespective of their bioassay conditions.…”

“…59 In this project, we relied on the HYPOGEN module of Discovery Studio soware suite (version 4.5) for supervised pharmacophore modelling. 60 Moreover, it was decided to use the "Common Feature Pharmacophore Generation" (also known as HIPHOP) protocol within Discovery Studio for unsupervised ligand-based pharmacophore modelling, i.e., to identify common binding features among potent ligands independent of their bioactivities, 59 which allowed us to incorporate additional training compounds irrespective of their bioassay conditions. The computational workow for ligandbased pharmacophore exploration is summarized in Fig.…”

“…Training compounds for unsupervised pharmacophore modelling need not to be bioassayed by the same procedure. 39,59,[62][63][64] Therefore, the pharmacophoric space of LSD-1 inhibitors was explored through 16 carefully selected training subsets (ESI Table SM7 †) from the collected inhibitors (1-281, bioassayed by a variety of procedures as in ESI Table SM1 †). Each subset was used to conduct 12 automatic runs (ESI Table SM8 †).…”

“…clinical practice until now. 21 The continued interest in LSD-1 prompted us to combine our innovative pharmacophore modelling methods 39,40,42,51,55,[59][60][61][62][63]65,66,69,73,74,76,96,101,105,106,108,109,112 with machine learning methodologies towards the discovery of new LSD-1 inhibitors. The fundamental novelty of the current project lies in allowing numerous pharmacophore models (of Fig.…”

Structure‐based Pharmacophore Screening Coupled with QSAR Analysis Identified Potent Natural‐product‐derived IRAK‐4 Inhibitors

“…[16] Compounds 1-71 (Table S1 in the Supplementary Material) were selected based on the fact that they were reported by the same research group and bioassayed using the same procedure, which is crucial to maintain statistical consistency. [17] A training set of 57 IRAK-4 inhibitors was selected for QSAR modeling, and the remaining 14 compounds (20 % of the dataset) were used as a testing set. The later set is essential to examine the predictive power of generated QSAR models.…”