SELFIES and the future of molecular string representations

Area of Science:

• Chemistry and Materials Science
• Artificial Intelligence
• Machine Learning

Background:

• Artificial intelligence (AI) and machine learning (ML) are increasingly used in chemistry and materials science for tasks like property prediction and molecular design.
• Traditional molecular string representations, such as Smiles, have limitations in AI/ML applications, often generating invalid chemical interpretations.
• The development of robust molecular representations is crucial for advancing AI in these scientific fields.

Purpose of the Study:

• To review the current landscape of molecular string representations for AI in chemistry.
• To highlight the advantages of the SELF-referencing embedded string (Selfies) language over traditional methods.
• To propose future research directions for robust molecular representations in AI-driven chemistry.

Main Methods:

• The study is a perspective piece, analyzing existing literature and proposing future work.
• It discusses the evolution and shortcomings of molecular string representations, focusing on Smiles and Selfies.
• Future projects are conceptualized based on identified challenges and opportunities.

Main Results:

• Selfies offers a 100% robust molecular string representation, overcoming the invalidity issues of Smiles.
• This new language has already enabled new applications in chemistry.
• The perspective outlines 16 concrete future projects for advancing robust molecular representations.

Conclusions:

• Robust molecular string representations are essential for the future of AI in chemistry and materials science.
• Selfies presents a significant advancement, offering greater reliability and enabling new possibilities.
• Further research into new chemical domains, AI/language interfaces, and interpretability is recommended to fully exploit these representations.