Exploring the Protein Sequence Space with Global Generative Models.

Area of Science:

• Biotechnology and Bioinformatics
• Computational Biology
• Artificial Intelligence

Background:

• Large-scale architectures for image and language processing have advanced computer vision and natural language processing (NLP).
• Breakthroughs in language models like ChatGPT and GPT-4 show exceptional capabilities in language tasks.
• These advancements are driving rapid development and unprecedented performance in protein research.

Purpose of the Study:

• To provide an overview of protein generative models.
• To review language models for novel artificial protein design.
• To explore non-transformer architectures and applications in directed evolution.

Main Methods:

• Review of recent literature on protein generative models.
• Analysis of language models applied to protein sequence generation.
• Examination of non-transformer architectures in protein design.
• Investigation of applications in directed evolution strategies.

Main Results:

• Language models are increasingly used for designing novel artificial proteins.
• Significant progress has been made in generating sequences in unexplored protein space.
• Various architectures, including transformers and non-transformers, are being employed.
• Generative models show promise in enhancing directed evolution approaches.

Conclusions:

• Generative models, particularly those based on language model architectures, are transforming protein design.
• The field is rapidly evolving with new methods achieving high performance.
• Future applications are expected in areas like synthetic biology and protein engineering.
• Continued research into diverse architectures and applications will further expand the potential of protein generative models.