Chemical language models for natural product discovery.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Natural products are a vital source of medicines, but their discovery is often slow and resource-intensive.
• Chemical language models (CLMs), processing string-based molecular representations, are emerging as powerful tools in natural product science.
• The field is rapidly evolving, with advancements from early neural networks to sophisticated large-scale Transformers.

Purpose of the Study:

• To provide a comprehensive overview of CLMs in natural product drug discovery.
• To trace the historical development and current applications of CLMs in this domain.
• To discuss challenges and future prospects of CLM-enabled natural product science.

Main Methods:

• Reviewing the literature on chemical language models and their application in natural product research.
• Analyzing the evolution of CLMs from foundational models to advanced Transformer architectures.
• Summarizing the predictive capabilities of CLMs for bioactivity, biosynthesis, and spectral data.

Main Results:

• CLMs significantly accelerate natural product discovery timelines.
• These models can predict key molecular properties and biosynthetic pathways.
• CLMs are instrumental in designing novel, natural-product-like scaffolds, expanding chemical space.
• Applications include predicting bioactivity, biosynthetic pathways, and spectral data.

Conclusions:

• CLMs are transforming natural product drug discovery, offering faster timelines and novel scaffold generation.
• Challenges such as data scarcity and model interpretability need to be addressed for further progress.
• The future of natural product science is increasingly intertwined with CLM advancements, promising new therapeutic agents.