Artificial Intelligence-Driven Inverse Design of Singlet Fission Candidates in the Acene family

In this work, we describe the successful application of an AI-driven inverse design protocol to identify singlet fission (SF) candidates within the acene family, exploring a chemical space of ≈10¹⁹ molecules. Substituent effects are encoded via Hammett σ constants, providing a chemically interpretable and generalizable descriptor for unseen functional groups. A Gated Recurrent Unit (GRU)-based Recurrent Neural Network (RNN) model is used to accurately predict excited singlet (S₁) and triplet (T₁) states energies. The model is coupled with optimization algorithms, including Genetic Algorithms (GA) and Particle Swarm Optimization (PSO), to efficiently navigate chemical space, optimize S₁ and T₁ energies, and reveal recurring substituent patterns. This approach not only identifies well-known tetracene and pentacene candidates but also uncovers viable benzene, naphthalene, and anthracene derivatives that satisfy SF criteria, which are typically inaccessible via intuition. The methodology, accessible at https://alba.ugr.es/acene/, establishes a versatile and interpretable platform for rational molecular design, enabling the exploration of large chemical spaces and the discovery of compounds with tailored excited-state properties.