Deep Learning Spectroscopy: Neural Networks for Molecular Excitation Spectra

Kunal Ghosh, Annika Stuke, Milica Todorović, Peter Bjørn Jørgensen, Mikkel N. Schmidt, Aki Vehtari, Patrick Rinke*

*Corresponding author for this work

Research output: Contribution to journalArticleScientificpeer-review

180 Citations (Scopus)
250 Downloads (Pure)


Deep learning methods for the prediction of molecular excitation spectra are presented. For the example of the electronic density of states of 132k organic molecules, three different neural network architectures: multilayer perceptron (MLP), convolutional neural network (CNN), and deep tensor neural network (DTNN) are trained and assessed. The inputs for the neural networks are the coordinates and charges of the constituent atoms of each molecule. Already, the MLP is able to learn spectra, but the root mean square error (RMSE) is still as high as 0.3 eV. The learning quality improves significantly for the CNN (RMSE = 0.23 eV) and reaches its best performance for the DTNN (RMSE = 0.19 eV). Both CNN and DTNN capture even small nuances in the spectral shape. In a showcase application of this method, the structures of 10k previously unseen organic molecules are scanned and instant spectra predictions are obtained to identify molecules for potential applications.

Original languageEnglish
Article number1801367
Pages (from-to)1-7
Number of pages7
JournalAdvanced Science
Issue number9
Publication statusPublished - 3 May 2019
MoE publication typeA1 Journal article-refereed


  • artificial intelligence
  • DFT calculations
  • excitation spectra
  • neural networks
  • organic molecules


Dive into the research topics of 'Deep Learning Spectroscopy: Neural Networks for Molecular Excitation Spectra'. Together they form a unique fingerprint.

Cite this