Generative design of singlet fission materials leveraging a fragment-oriented database.

Area of Science:

• Materials Science
• Computational Chemistry
• Organic Chemistry

Background:

• Fragment-based high-throughput virtual screening (HTVS) and genetic algorithms (GA) were used to identify singlet fission (SF) molecular candidates from the FORMED repository.
• Previous methods identified known SF chromophores like acenes and boron-dipyrromethane (BODIPY), but were limited by predefined fragments, restricting exploration of novel molecular cores.

Purpose of the Study:

• To train a generative learning framework using reinforcement learning and property predictions to explore a wider chemical space for SF chromophore discovery.
• To identify novel molecular scaffolds beyond derivatives of known chemistry for tailored material applications.

Main Methods:

• Leveraged the FORMED repository (116,687 molecules) to train a generative learning framework.
• Employed reinforcement learning and property predictions to guide the generative model.
• Conducted in-depth investigation of predicted candidates, including electronic structure analysis.

Main Results:

• The generative model successfully rediscovered known SF chromophore classes such as polyenes, benzofurans, fulvenoids, and quinoidal systems.
• An unexpected neocoumarin scaffold (2-benzopyran-3-one) was identified, absent from the training data.
• The neocoumarin scaffold exhibited diradicaloid behavior comparable to known SF compounds like 2-benzofuran.

Conclusions:

• Generative models combined with property prediction offer a powerful approach to discover novel molecular candidates beyond existing chemical knowledge.
• This methodology enables the exploration of new chemical space for designing materials with specific properties, such as singlet fission.
• The discovery of the neocoumarin scaffold highlights the potential for finding unforeseen molecular cores for advanced material applications.