Skip to main content
. 2023 Jun 23;6:132. doi: 10.1038/s42004-023-00932-3

Fig. 4. Spec2Mol architecture.

Fig. 4

The Spec2Mol model consists of a spectra encoder and a SMILES decoder which have been trained separately but share the same embedding space. a The AE is pre-trained to translate from a random SMILES to the canonical SMILES string. b The spectra encoder is trained to learn the same embedding as the SMILES encoder. c During inference, the spectra encoder and the SMILES decoder of the pre-trained model are used to translate spectra into molecular structures.