Deep-Learning Generative Design of Organic Scintillator Molecules for Sub-GeV Dark Matter Detection
A proof-of-concept pipeline that pairs a variational autoencoder (which generates novel organic molecules from their string encodings via a continuous latent space) with a multilayer-perceptron property predictor, trained on the PubChemQC dataset, to design and screen candidate molecular scintillators optimised for sub-GeV dark matter direct detection. The network targets low (few-eV) electronic transition energies and large oscillator strengths as proxies for detector energy threshold and quantum efficiency, generating molecules absent from even the largest quantum-chemistry databases and using clustering to surface the most promising structures without expensive DFT calculations.
Deep-Learning Generative Design of Organic Scintillator Molecules for Sub-GeV Dark Matter Detection
Direct searches for sub-GeV dark matter are limited by the intrinsic quantum properties of the target material. In this proof-of-concept study, we argue that this problem is particularly well suited …