A simplified machine learning workflow for identifying potential singlet fission candidates: benzannulated biphenylenes as a case study

Abstract

Singlet fission (SF) is a physical phenomenon exhibited by some families of organic materials potentially able to boost the power conversion efficiency of solar cells beyond the theoretical Shockley–Queisser limit of 33%. To experience SF, a molecule must fulfil the so-called energy matching conditions (EMCs), which can be evaluated using DFT and TD-DFT calculations. Here, we propose a simple protocol that exploits machine learning workflows to screen large libraries of molecules using a reduced number of quantum chemical calculations. The protocol is based on the AQME and ROBERT platforms and is adapted to users with no experience in data science and basic computational chemistry knowledge. Using this approach, we screened a library of 3835 benzannulated biphenylenes to identify 505 candidates fulfilling the first EMC for SF. Fragment-based statistical analysis was employed to rationalize the structural features associated with SF. The workflow is general and can be applied to other families of compounds for the accelerated discovery of SF materials.