Machine learning-based 3D-QSAR models for predicting the estrogen receptor-binding activity of small molecules
BR. Bharath, Sreerupa Mitra, Nadeem Ahmad Khan · In Silico Research in Biomedicine · 2025
Estrogen receptor alpha (ERα) belongs to the steroid receptor superfamily and acts as a ligand-activated transcription factor. Structurally, ERα comprises six domains labeled A through F. The DNA-binding domain (DBD) facilitates specific interactions with estrogen response elements, while the ligand-binding domain (LBD) engages with various agonistic and antagonistic hormones. Notably, numerous endocrine-disrupting chemicals (EDCs) exert adverse effects on estrogen signaling by interacting with ERα. Consequently, there is a critical need to evaluate the endocrine disruption potential of new chemical entities (NCEs) by scrutinizing their interactions with multiple pertinent targets. ERα is a pivotal target protein, and leveraging third-party tools for predicting the relative binding affinity (RBA) of small molecules via 2D-QSAR has become common. However, in this study, we advanced beyond conventional methods by developing machine learning-based 3D-QSAR models. These 3D-QSAR models were built using the classification dataset of VEGA V.1.2.0 for the estrogen receptor IRFMN-CERAPP and IRFMN-RBA models. Our investigation demonstrated that the 3D-QSAR models, which employ algorithms such as random forest (RF), support vector machine (SVM), and multilayer perceptron (MLP), outperform the VEGA models in terms of accuracy, sensitivity, and selectivity. Furthermore, the efficacy of these models was corroborated through validation against external datasets. Notably, the 3D-QSAR models exhibit superior accuracy and sensitivity compared to the VEGA model, thereby offering a promising approach for assessing the endocrine disruption potential of novel chemical entities.