Scalable Peptide Design via Memory-Efficient Equivariant Transformer

Rui Jiao *, Xiangzhe Kong *, Yinjun Jia *, Yijia Zhang, Ziyi Yang, Yang Liu, Jianzhu Ma † · 2026

Read Scalable Peptide Design via Memory-Efficient Equivariant Transformer by Rui Jiao *, Xiangzhe Kong *, Yinjun Jia *, Yijia Zhang, Ziyi Yang, Yang Liu, Jianzhu Ma † (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-25006.txt.

Description

Rui Jiao and six colleagues introduce Meet, a memory efficient transformer for designing peptides atom by atom. Peptide design needs sequence and structure together, under full atom level constraints. Latent generative methods shrink atomic structures into block level codes. Their scaling, though, depends on the backbone doing the encoding, decoding, and denoising. Meet is an E(3) equivariant backbone keeping two coupled feature streams, scalar and vector.

Its core work is rebuilt around memory efficient attention. Vector features start from a global pooling of coordinates. Pairwise distances enter through augmented query and key products, and bond information through sparse bond adaptation. Plugged into an autoencoder and latent diffusion pipeline, Meet uses memory that grows linearly with atom count. Output quality improves over peptide design methods. On datasets derived from the AlphaFold database, the backbone supports scaling of model and data, giving better physical validity, binding affinity, and sample diversity.

Catalog record

CREATORRui Jiao *, Xiangzhe Kong *, Yinjun Jia *, Yijia Zhang, Ziyi Yang, Yang Liu, Jianzhu Ma †
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2606.25006

Browse the full library · About the Tower of Babel