Variable-Length Protein Design with GPFlow
Our new work, Variable-Length Generative Protein Design via Generalized Poisson Flow, introduces GPFlow, a unified framework that learns protein length jointly with structure and sequence instead of fixing it before sampling.
We have made variable-length generative modeling work for protein design. Not on a single task, but 𝟝 across structure/sequence design, their corresponding motif scaffolding, and peptide codesign. Not on a single modality, but continuous, discrete, Riemannian, and their mixes. pic.twitter.com/LpUBY5xx4N
— Chaoran Cheng (@ccr_cheng) July 13, 2026