Extrapolative Controlled Sequence Generation via Iterative Refinement
This addresses a critical challenge in automated design like drug discovery, where generating novel, improved sequences is essential, though it builds incrementally on iterative refinement methods.
The paper tackles the problem of generating sequences with attribute values beyond the training distribution, such as designing more stable proteins, by proposing Iterative Controlled Extrapolation (ICE), which iteratively edits sequences to enable extrapolation. Results show ICE considerably outperforms state-of-the-art approaches on protein engineering and natural language tasks.
We study the problem of extrapolative controlled generation, i.e., generating sequences with attribute values beyond the range seen in training. This task is of significant importance in automated design, especially drug discovery, where the goal is to design novel proteins that are \textit{better} (e.g., more stable) than existing sequences. Thus, by definition, the target sequences and their attribute values are out of the training distribution, posing challenges to existing methods that aim to directly generate the target sequence. Instead, in this work, we propose Iterative Controlled Extrapolation (ICE) which iteratively makes local edits to a sequence to enable extrapolation. We train the model on synthetically generated sequence pairs that demonstrate small improvement in the attribute value. Results on one natural language task (sentiment analysis) and two protein engineering tasks (ACE2 stability and AAV fitness) show that ICE considerably outperforms state-of-the-art approaches despite its simplicity. Our code and models are available at: https://github.com/vishakhpk/iter-extrapolation.