ncbi/CellPuzzles

Cell-o1是一个用于训练大型语言模型解决单细胞推理谜题的数据集，通过强化学习进行训练。数据集包含来自同一捐赠者的细胞批次，每个细胞需要从共享的候选集中分配一个唯一的类型，以确保在批次级别的全局一致性。数据集分为训练集、测试集和推理集，适用于细胞类型注释任务，特别是在分析单细胞RNA测序数据的异质性方面。

Cell-o1 is a dataset for training large language models to solve single-cell reasoning puzzles through reinforcement learning. The dataset contains batches of cells from the same donor, with each cell needing to be assigned a unique type from a shared candidate set to ensure global consistency at the batch level. The dataset is split into training, test, and reasoning sets, suitable for cell type annotation tasks, especially in analyzing the heterogeneity of single-cell RNA sequencing data.