Computational Prediction of Endocrine-Disrupting Xenobiotics Associated with Hormone-Related Cancer Classification Using Molecular and Physicochemical Features

Background: Chemicals associated with hormone-related cancer may interfere with endocrine signaling and are linked to biological processes involved in cancer development. Conventional toxicity assessment relies heavily on laboratory experiments and animal studies. Existing computational approaches are also limited by a lack of validated toxicity datasets and challenges in modeling complex biological mechanisms. Methods: This study presents an artificial intelligence-based framework for the binary classification of chemicals associated with hormone-related cancer. Chemical information was collected from PubChem and combined with biological information from the Comparative Toxicogenomics Database (CTD). CTD disease association records were used to construct the binary Hormone_Cancer_Label, while the disease association information was used for label generation. RDKit was used to process SMILES representations and generate molecular descriptors, while character-level TF-IDF features were extracted from the canonical SMILES. Results: The resulting Toxicity Cancer dataset contained 10,000 unique chemicals and was stratified into training (7000; 70%), validation (1500; 15%), and testing (1500; 15%) sets. Several machine learning and deep learning models, including the proposed CNN-BiLSTM architecture, were evaluated. The proposed model achieved an accuracy of 99.07%, a precision of 98.94%, a recall of 99.20%, an F1-score of 99.07%, and an ROC-AUC of 98.18%. Conclusions: The framework provides a computational classification of chemicals associated with hormone-related cancer and supports early-stage hazard screening and prioritization for toxicological investigation.