Reproduce “Controllable Protein Sequence Generation with LLM Preference Optimization” 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
Xiangyu Liu, Yi Liu, Silei Chen, Wei Hu
A new method, CtrlProt, improves controllable protein sequence generation by employing a multi-listwise preference optimization strategy that explicitly incorporates structural stability and functional relevance metrics. It generates novel protein sequences with enhanced stability and functionality for both single and multiple attributes, even with limited multi-attribute data, outperforming existing LLM-based baselines.
Key insights
- Explicitly incorporating structural stability and functionality metrics into the preference optimization process is critical for generating biologically viable and robust protein sequences.
- A multi-listwise preference optimization strategy, leveraging ranking and quality difference information, can effectively refine protein generation quality and overcome limitations of standard DPO.
- Multi-attribute protein design can be achieved by combining single-attribute representations and applying preference optimization, significantly mitigating the impact of data scarcity for complex designs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.