DeepTargetClass: a web-based platform for predicting protein target classes of small molecules.

The identification of protein target classes is a key step in drug discovery, as it enables prioritization of screening campaigns and supports target-based drug repurpose. In this study, we developed a deep-learning pipeline based on a multilayer perceptron (MLP) trained on 15,804 curated compounds representing four major pharmacological target classes: G protein-coupled receptors (GPCRs), kinases, nuclear receptors, and transporters. Using extended connectivity fingerprints (ECFP4) as molecular descriptors, the model achieved 96% accuracy in internal cross-validation and 87% accuracy on an external test set, demonstrating performance comparable to ensemble classifiers such as Random Forest, XGBoost, and LightGBM. Class-specific F1 scores confirmed robust and balanced predictions across GPCR, kinase, nuclear receptor, and transporter categories. Model interpretability was addressed using SHAP values, which highlighted pharmacophore-like substructures consistent with known ligand-target interactions. Application to reference drugs further validated predictive utility, with correct assignment of most compounds to their canonical protein target class. The final MLP model was deployed as a user-friendly web application to facilitate accessible protein class prediction for novel compounds. Overall, this work presents a reliable and interpretable computational framework to support target-class-based drug discovery and repositioning.