Attribution assignment for deep-generative sequence models enables interpretability analysis using positive-only data
Robert Frank, Michael Widrich, Rahmad Akbar, Günter Klambauer, Geir Kjetil Sandve, Philippe A. Robert, Victor Greiff, Pingzhao Hu · 2026
Read Attribution assignment for deep-generative sequence models enables interpretability analysis using positive-only data by Robert Frank, Michael Widrich, Rahmad Akbar, Günter Klambauer, Geir Kjetil Sandve, Philippe A. Robert, Victor Greiff, Pingzhao Hu (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/PMC13638040.txt.
Description
Robert Frank and seven colleagues introduce GAMA, a method for asking generative AI models which parts of a biological sequence drove their designs. Generative models learn from successful examples only, such as antibodies that bind well, which suits biology because failure data are scarce or meaningless. Their opacity was the problem: no attribution tools existed for them. GAMA adapts a gradient-based technique, comparing a trained model's internal signals against an untrained reference.
Synthetic datasets with known answers let the team check that the method recovers genuinely important positions and behaves statistically as it should. Applied to antibody-antigen binding data, GAMA surfaced residues that match biological knowledge, highlighting key positions in real experimental sets. Because it needs no negative examples, the method fits exactly the settings where generative design is most used. The authors offer it as a way to audit black-box protein design before trusting its outputs.
Catalog record
| CREATOR | Robert Frank, Michael Widrich, Rahmad Akbar, Günter Klambauer, Geir Kjetil Sandve, Philippe A. Robert, Victor Greiff, Pingzhao Hu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |