Insight Analysis of Promiscuous Estrogen Receptor α-Ligand Binding by a Novel Machine Learning Scheme

Abstract

Estrogen receptor α (ERα) plays a significant role in occurrence of breast cancer and may cause various adverse side-effects when ERα is an off-target protein. A theoretical model was derived to predict the binding affinity of ERα using the pharmacophore ensemble/support vector machine (PhE/SVM) scheme to consider the promiscuous characteristic of ERα. The estimations by PhE/SVM were discovered to be in good agreement with the observed values for those training molecules ( n = 31, r = 0.80, q = 0.77, RMSE = 0.57, s = 0.58), test molecules ( n = 179, q = 0.91-0.96, RMSE = 0.33, s = 0.26) and outliers ( n = 15, q = 0.80-0.86, RMSE = 0.56, s = 0.49). When subjected to various statistical validations, the PhE/SVM model consistently fulfilled the strictest criteria. A mock test also asserted its predictivity. When compared with crystal structures, the calculated results are consistent with the reported ERα-ligand co-complex structure, and the plasticity nature of ERα is also disclosed. Consequently, this precise, fast, and robust model can be adopted to predict ERα-ligand binding affinities and to design safer non-ERα-targeted pharmaceuticals in the process of drug discovery and development.

Towards accurate high-throughput ligand affinity prediction by exploiting structural ensembles, docking metrics and ligand similarity

“…Our work confirmed the performance of Random Forest over other machine learning approaches as previously noticed (Russo et al , 2018). In some cases, higher accuracy was reported but for smaller compound libraries (Hou et al , 2018). Accordingly, our results present one of the largest validation surveys and best performing tools for affinity prediction against ER α .…”

“…Thus, a better understanding of the mechanism of ligand recognition by ER α is of paramount importance for safer drug design. Previously, dedicated prediction methods have been addressing the question of whether a molecule is binding or not (Niu et al , 2016; Pinto et al , 2016; Ribay et al , 2016; Mansouri et al , 2016), and traditional structure-activity relationship (QSAR) modeling studies have been also performed with varying success on this nuclear receptor (Waller et al , 1995; Waller, 2004; Asikainen et al , 2004; Zhang et al , 2013; Zhao et al , 2017; Hou et al , 2018).…”

Exploring Ligand Binding Domain Dynamics in the NRs Superfamily

“…The analyses carried out here allow the rapid identification of the most flexible residues and, in particular, the fast and easy identification of reference structures to be used in virtual screening campaigns for any other NRs sufficiently represented, without the need of running more extensive and demanding calculations [ 70 ]. The presented methodology, in fact, only relies on the comparison of pockets based on the ID of the lining residues without paying attention to their chemical characteristics and making the calculation quite fast.…”

Prediction of the Antioxidant Response Elements' Response of Compound by Deep Learning

“…Recently, deep learning (Lecun et al, 2015), as a promising machine learning method, has been applied in a wide range of fields, such as physics, life science and medical science (Gulshan et al, 2016). There were also some researches in biology (Mamoshina et al, 2016; Dang et al, 2018; Hou et al, 2018) and drug design areas (Gawehn et al, 2016; Hughes and Swamidass, 2017). Furthermore, deep learning methods have been also applied in small molecule toxicity assessment (Blomme and Will, 2016).…”