We trained a model to co-generate protein sequence and structure by working in the ESMFold latent space, which encodes both. PLAID only requires sequences for training but generates all-atom structures!

Really proud of @amyxlu.bsky.social 's effort leading this project end-to-end!