Posted: Tuesday, August 15, 2023
According to findings presented in the Journal of Cheminformatics, a novel deep learning–based hybrid framework known as DeepAR may be able to identify potential androgen receptor (AR) antagonists accurately and quickly. Therapies targeting androgen receptors often lead to treatment resistance and disease progression in prostate cancer; the identification of new antagonists may help combat such resistance and improve treatment outcomes.
“We anticipate that DeepAR could be a useful computational tool for community-wide facilitation of [androgen receptor] candidates from a large number of uncharacterized compounds,” concluded Watshara Shoombuatong, PhD, of Mahidol University, Bangkok, Thailand, and colleagues.
DeepAR is capable of efficiently identifying potential AR antagonists using only the Simplified Molecular Input Line Entry System (SMILES) notation, which translates chemicals’ structures into code comprehensible to certain software. The foundation for the machine learning framework was a benchmark data set comprising compounds from the chEMBL database categorized by activity against androgen receptors. Based on these data, several commonly known molecular descriptors and machine learning algorithms were then used to develop various baseline models. A combination of probabilistic features created by these models were employed to establish a meta-model referencing a unidimensional convolutional neural network.
Results from this meta-model suggested that DeepAR’s approach to antagonist identification was accurate and stable. An additional computational approach known as Shapley Additive Explanations (SHAP) allowed the model to provide feature importance information, and SHAP’s waterfall plot and molecular docking resulted in the description and analysis of potential antagonists. Some notable characteristics of antagonist candidates included N-heterocyclic moieties, halogenated substituents, and a cyano functional group.
“A webserver for our proposed model DeepAR has been constructed to provide the scientific community with a practical tool that can be widely used for performing high-throughput identification of AR antagonists in an economic manner,” noted the authors.
Disclosure: The study authors reported no conflicts of interest.