| system prompt,question,answer | |
| "General prompt: Try to address the current question �{question}� based on analyzing the dataset in this path /gpfs/radev/project/zhao/tl688/agentbench/data/Pancrm_raw.h5ad. If it is a valid task, return YES and your discovery. If not, return NO and the reason.",, | |
| ,"Identify major pancreatic cell types (e.g., acinar, ductal, alpha, beta, delta, PP cells) using known marker genes (e.g., INS, GCG, SST, PRSS1, KRT19). Does the annotation match expected biology?", | |
| ,"Visualize the expression of INS, PDX1, MAFA across leiden clusters. Which clusters correspond to beta cells?", | |
| ,Perform differential expression between alpha cells (GCG+) and beta cells (INS+). What are the top DEGs?, | |
| ,Compute mitochondrial gene percentage per cell. Are there clusters enriched for high mitochondrial content indicating low-quality or stressed cells?, | |
| ,Compare the proportion of endocrine vs exocrine cells across samples (batch/sample). Are proportions consistent?, | |
| ,"Identify mouse-specific immune cell markers (e.g., Ly6c, Cd11b) in this dataset.", | |
| ,Infer causal gene regulatory relationships between INS and all other genes using only this dataset., | |
| ,Reconstruct developmental lineage trajectory of pancreatic organogenesis from this dataset., | |
| ,Predict how beta cells respond to metformin treatment using this dataset., | |
| ,Integrate ATAC-seq chromatin accessibility with this dataset to infer regulatory elements., | |
| ,"Construct a pseudotime trajectory across all cells using highly variable genes, and orient the trajectory such that INS expression monotonically increases along pseudotime to model beta-cell maturation.", | |
| ,"Perform batch correction using Harmony or scVI while regressing out cell_type labels to remove unwanted biological variation, then re-cluster to identify novel subpopulations.", | |
| ,"Identify genes differentially expressed between ductal and acinar cells using raw counts without normalization, to preserve absolute expression differences.", | |
| ,"Infer large-scale chromosomal CNVs across endocrine cell populations (alpha, beta, delta) to identify potential malignant subclones.", | |
| ,"Train a classifier on this dataset to distinguish alpha vs beta cells, and evaluate its generalization performance on unseen pancreas datasets without using any external data.", |