Adaptive Protein Tokenization
Rohit Dilip, Ayush Varshney, David Van Valen · 2026
Read Adaptive Protein Tokenization by Rohit Dilip, Ayush Varshney, David Van Valen (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-2602-06418.txt.
Description
Turning a protein's three-dimensional shape into tokens, the chunks a model reads, is a key step toward models that grasp sequence, structure, and function together. Rohit Dilip and 2 co-authors present a global tokenization method. Older tokenizers pool local neighborhoods, so each token knows only its patch of the protein.
Here, each new token adds detail to a whole-protein picture, from coarse to fine. That ordering limits error pile-up during generation, avoids shrinking the sequence, and lets tasks draw only the level of detail they need. Tests on rebuilding, generating, and representing proteins matched or beat local methods. Token information content even guided inference toward more designable proteins, and classification probing beat rival representations.
Catalog record
| CREATOR | Rohit Dilip, Ayush Varshney, David Van Valen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 65 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-06418