BitterTranslate: A Natural Language Processing and Machine Learning-based Framework for Mapping Bitter Taste Receptor Agonism
Bibliographic record
Abstract
Bitter taste receptors (TAS2Rs) are G protein-coupled receptors (GPCRs) expressed on the tongue and by many extraoral tissues. Identifying TAS2R ligands is an area of interest for improving bitter-drug compliance, treating various illnesses, and studying the receptors’ extraoral functions. Although machine learning, emerging as a promising tool for drug discovery, can in theory be used for predicting TAS2R activators, obtaining high-quality features from which the machine learning model can learn is time-intensive and reliant on specialized software. This work explores the potential of transformers (a neural network architecture that has revolutionized natural language processing-based tasks and is a powerful tool for extracting features from sequential biomolecular and chemical data) in computer-aided drug design, specifically for predicting potential ligands for TAS2R activation. We developed BitterTranslate, a screening algorithm to predict TAS2R agonists based solely on a Simplified Molecular-Input Line-Entry System (SMILES) string of the ligand and the amino acid sequence of the TAS2R. Bidirectional Encoder Representations from Transformers (BERT) models trained on small molecules and GPCRs were used to extract ligand and receptor features. An XGBoost classifier was pre-trained on a large GPCR–ligand dataset and fine-tuned on the smaller TAS2R–ligand dataset. The algorithm predicts ligand associations with an 80% precision and 65% recall across all TAS2Rs and an 83% precision and 88% recall for the top receptor, TAS2R14. Since BitterTranslate performs reasonably well for TAS2Rs for which the data is scarce, it can be expected to perform even better for other more populated families of GPCRs with more ligand information available.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".