Reproduce “Generative modeling of intrinsically disordered protein regions by reinforcing sparse autoencoder features” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jason X. Liu, Sebastian Ibarraran, Frank Hu, Soojung Yang, Xinyu A. Feng, Abigail Park, Anagha Aneesh, Lacramioara Bintu and 2 more
IDiom is an autoregressive model trained on predicted intrinsically disordered regions. RL-SAE post-trains it to generate sequences activating selected sparse-autoencoder features, rewarding coverage of patterns rather than steering toward them. Across eight design tasks, it activates an average of 90% of 30 targeted features, versus 24% with steering. Predicted activity and localization readouts provide computational evidence, not experimental validation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.