Application of Transformers in Cheminformatics.

Area of Science:

• Computational chemistry
• Machine learning
• Artificial intelligence in chemistry

Background:

• Computing accelerates chemical processes, with machine learning (ML) identifying patterns for tasks like property prediction and substance generation.
• Complex chemical data necessitates advanced ML architectures, with transformer models showing significant promise.

Purpose of the Study:

• To review recent advancements in applying transformer architectures to machine learning problems in chemistry.
• To provide a comprehensive summary of existing works adapting transformers for chemical data analysis.

Main Methods:

• Structured review of publications on transformer architectures in chemistry.
• Analysis based on diverse chemical data representations.
• Highlighting strengths and weaknesses of different representations and models.

Main Results:

• Transformers are increasingly adopted in chemistry, revolutionizing ML applications.
• Adaptation of transformers varies across different chemical data representations.
• Ongoing research explores diverse use cases and model architectures.

Conclusions:

• Transformer models offer powerful solutions for complex chemical learning tasks.
• The choice of chemical representation is crucial for effective transformer model application.
• Future directions involve further exploration of transformer capabilities across the chemical domain.