Spectra to structure: contrastive learning framework for library ranking and generating molecular structures for infrared spectra

Inferring complete molecular structure from infrared (IR) spectra is a challenging task. In this work, we propose SMEN (Spectra and Molecule Encoder Network), a framework for scoring molecules against given IR spectra. The proposed framework uses contrastive optimization to obtain similar embedding for a molecule and its spectra. For this study, we consider the QM9 dataset with molecules consisting of less than 9 heavy atoms and obtain simulated spectra. Using the proposed method, we can rank the molecules using embedding similarity and obtain a Top 1 accuracy of similar to 81%, Top 3 accuracy of similar to 96%, and Top 10 accuracy of similar to 99% on the evaluation set. We extend SMEN to build a generative transformer for a direct molecule prediction from IR spectra. The proposed method can significantly help molecule library ranking tasks and aid the problem of inferring molecular structures from spectra.