Fig. 4: Difficulty in learning chirality for Transformer. | Nature Communications

Fig. 4: Difficulty in learning chirality for Transformer.

From: Difficulty in chirality recognition for Transformer architectures learning chemical structures from string representations

Fig. 4

a, b Temporal change of perfect accuracy when each one of the characters in Simplified Molecular Input Line Entry System (SMILES) was masked for trainings in which stagnation did/did not occur. Rare tokens which did not appear in the test set are not shown. c Examples of target and predicted molecules during stagnation (at step 10,000). Each of the molecules in the upper row is predicted targeted to the directly below molecule. d Ratio of correct predictions, predictions with only mistakes of “@” token for “@@” token and “@@” token for “@” token (mistakes attributed to chirality), and predictions with other mistakes in the test set. Source data are provided as a Source Data file.

Back to article page