Accelerating Life Sciences Research

OpenAI and Retro Biosciences are accelerating life sciences research by developing GPT-4b micro, a small model specialized for protein engineering. This model is based on a scaled-down version of GPT-4o, and was trained by combining protein sequences, biological text, and 3D structural data.

Using this model, they designed variants of Yamanaka factors, a Nobel Prize-winning research topic, successfully increasing stem cell reprogramming marker expression by more than 50x compared to before. The designed proteins also showed improved DNA damage repair ability, demonstrating higher rejuvenation potential, and their effects were validated across various cell types and delivery methods.

The key features of GPT-4b micro are as follows:

• Can design proteins incorporating their functional context (text descriptions, co-evolutionary homologous sequences, etc.)
• Effectively handles proteins with intrinsically disordered regions whose structure is not fixed
• Supports a context length of up to 64,000 tokens, which is exceptional for a protein model