| Label | Score | |
|---|---|---|
| 😋 | Sweet | 0.9952 |
| 😝 | Bitter | 0.0021 |
| ❓ | Undefined | 0.0015 |
| 😖 | Sour | 0.0010 |
| 🤤 | Umami | 0.0002 |
| SMILES Drawing | Similarity Score | Flavor |
|---|---|---|
| 1.0000 | sweet | |
| 0.9714 | sweet | |
| 0.9510 | sweet |
Determining the taste of molecules is a labor and time intensive process in food chemistry. Various machine learning approaches have been used for taste classification. However, despite their ubiquity in sequential learning tasks, transformer models have not been tested for taste classification so far.
A given query molecule is checked against the database, and the three closest matching molecules (based on Tanimoto distance) are returned - these are shown under "Similar Compounds". Additionally, FART predicts the cross-entropy for all five categories to classify it.