PLAID: Generating Proteins with Latent Diffusion and Protein Folding Models
Published:Apr 8, 2025 10:30
•1 min read
•Berkeley AI
Analysis
This article introduces PLAID, a novel multimodal generative model that leverages the latent space of protein folding models to simultaneously generate protein sequences and 3D structures. The key innovation lies in addressing the multimodal co-generation problem, which involves generating both discrete sequence data and continuous structural coordinates. This approach overcomes limitations of previous models, such as the inability to generate all-atom structures directly. The model's ability to accept compositional function and organism prompts, coupled with its trainability on large sequence databases, positions it as a promising tool for real-world applications like drug design. The article highlights the importance of moving beyond structure prediction towards practical applications.
Key Takeaways
- •PLAID addresses the multimodal co-generation problem in protein design.
- •The model can be trained on large sequence databases.
- •It aims to bridge the gap between structure prediction and real-world applications.
Reference
“In PLAID, we develop a method that learns to sample from the latent space of protein folding models to generate new proteins.”