Reproduce “HarmonyCell: Automating Single-Cell Perturbation Modeling under Semantic and Distribution Shifts” 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
Wenxuan Huang, Mingyu Tsoi, Yanhao Huang, Xinjie Mao, Xue Xia, Hao Wu, Jiaqi Wei, Yuejin Yang and 5 more
HarmonyCell, an agent framework from Fudan University and Shanghai Artificial Intelligence Laboratory, automates single-cell perturbation modeling by integrating an LLM-driven semantic unifier and an adaptive Monte Carlo Tree Search engine. This approach achieved a 95% valid execution rate on heterogeneous single-cell datasets and improved generalization performance, often outperforming specialized baselines on various perturbation prediction tasks.
Key insights
- An LLM-driven approach effectively resolves semantic heterogeneity, allowing robust data ingestion and processing for uncurated single-cell datasets where general coding agents fail.
- The hierarchical action space within MCTS is crucial for efficiently exploring and identifying optimal model paradigms and architectures, avoiding local optima in complex model design tasks.
- Automated data unification not only resolves syntactic conflicts but also facilitates positive transfer of biological signals, enabling improved generalization through scalable data integration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.