BitterTranslate: A Natural Language Processing and Machine Learning-Based Framework for Mapping Bitter Taste Receptor Agonism

Bitter taste receptors (TAS2Rs) are G protein-coupled receptors (GPCRs) expressed most notably on the tongue, but also in many extraoral tissues. TAS2R ligands are used for improving bitter-drug compliance, treating various illnesses, and studying the receptors' extraoral functions. Although machine learning, a promising drug discovery tool, can be used to predict TAS2R activators, obtaining high-quality features for training is time-intensive and reliant on specialized software. This work explores the potential of transformers (a neural network architecture for extracting features from sequential text strings that has revolutionized natural language processing-based tasks) for predicting these ligands. Hence, BitterTranslate, a TAS2R-agonist prediction algorithm, needs only the Simplified Molecular-Input Line-Entry System (SMILES) string of the ligand and the amino acid sequence of the TAS2R. The algorithm was built using two Bidirectional Encoder Representations from Transformers (BERT) models: one trained on small molecules to extract ligand features and the other trained on GPCRs to extract receptor features. An XGBoost classifier was pretrained on a large GPCR-ligand data set and fine-tuned on the smaller TAS2R-ligand data set. BitterTranslate predicts ligand associations with 80% precision and 65% recall across all TAS2Rs and 83% precision and 88% recall for the receptor with the most data: TAS2R14.