AndroPred: an artificial intelligence-based model for predicting androgen receptor inhibitors

AbstractAndrogen receptor (AR), a steroid receptor, plays a pivotal role in the pathogenesis of prostate cancer (PCa). AR controls the transcription of genes that help cells avoid apoptosis and proliferate, thereby contributing to the development of PCa. Understanding AR molecular mechanisms has led to the development of newer drugs that inhibit androgen production enzymes or block ARs. The FDA has approved a small number of AR-inhibiting drugs for use in PCa thus far, as the identification of novel AR inhibitors is difficult, expensive, time-consuming, and labor-intensive. To accelerate the process, artificial intelligence (AI) algorithms were employed to predict AR inhibitors using a dataset of 2242 compounds. Four machine learning (ML) and deep learning (DL) algorithms were used to train different prediction models based on molecular descriptors (1D, 2D, and molecular fingerprints). The DL-based prediction model outperformed the other trained models with accuracies of 92.18% and 93.05% on the training and test datasets, respectively. Our findings highlight the potential of DL, particularly the DNN model, as an effective approach for predicting AR inhibitors, which could significantly streamline the process of identifying novel AR inhibitors in PCa drug discovery. Further validation of these models using experimental assays and prospective testing of newly designed compounds would be valuable to confirm their predictive power and applicability in practical drug discovery settings.Communicated by Ramaswamy H. SarmaKeywords: Androgen receptor (AR)inhibitorsdeep neural network (DNN)k-nearest neighbor (kNN)machine learning (ML)random Forest (RF)support vector machine (SVM) Disclosure statementNo potential conflict of interest was reported by the author(s).Data and software availabilityThe AndroPred program is available at https://github.com/PGlab-NIPER/AndroPred.git.Supporting informationThe dataset used to develop the AndroPred model for the prediction of androgen receptor inhibitors (Supplementary Table S1)Top 100 features selected by the Boruta, recursive feature elimination (RFE) and random forest (RF) algorithms (Supplementary Table S2).Features used to train machine learning (ML)-based prediction models (Supplementary Table S3).Results of cut point analysis of the top 50 features (Supplementary Table S4).Data availability statementThe developed model is available at https://github.com/PGlab-NIPER/AndroPred.gitAdditional informationFundingThis work has been supported by the Department of Biotechnology (DBT Project BT/PR40164/BTIS/137/17/2021).