Unsupervised discovery of functional sequence patterns from protein language model with MotifAE
Chao Hou, Di Liu, Yufeng Shen · 2026
Read Unsupervised discovery of functional sequence patterns from protein language model with MotifAE by Chao Hou, Di Liu, Yufeng Shen (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/pmc-13631247.txt.
Description
Chao Hou, Di Liu, and Yufeng Shen fish readable patterns out of protein language models, in a 2026 paper. These models absorb evolutionary regularities from millions of sequences, but the patterns sit locked inside dense number vectors. The authors wanted them in a form humans can inspect.
Their tool, MotifAE, passes model embeddings through a sparse autoencoder, a bottleneck that forces a few interpretable features to carry the signal. An extra smoothness penalty keeps features coherent along the sequence, sharply improving recovery of known functional motifs. The discovered patterns are diverse, match established motifs, and echo in the model's own weights. Some features capture whole structural domains, and their activation tracks which residues matter for a domain's job. One feature set predicted how mutations affect protein stability. The authors offer the framework for function analysis and protein engineering.
Catalog record
| CREATOR | Chao Hou, Di Liu, Yufeng Shen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 75 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |