Bring Chemical Intuition to Chips: Transferable Chemical-intuitive Model to Predict Photophysics of Organic Aggregates

While machine-learning methods indicated good adaptability for machine-learning algorithms in the pan-chemistry field by its breakthroughs in pharmacy. Materials research still benefits from such new techniques fewer due to the inconsistency in the paradigm of study in the diversely different subareas which demand special treatment individually. In this contribution, we proposed an innovative design of the embedding method, which is inspired by chemical intuition, to bring neural networks into the field for modelling photophysics of the organic light-emitting materials in condensed states. We outline this framework and demonstrate its successful implementation in the predictive classification of fluorophores by its mechanisms, direction of spectra shift from solution to solid-state, and regression of spectral features, including emission peaks wavelengths in pristine solid-state and nano-aggregates. Our work may serve as an example for a specific area of materials research to transfer the empirical chemical intuition into machine-learning models and build comprehensive performance-oriented pre-screening systems to develop new compounds with demanded characters.