SHAPE (Self-supervised morpHology Agent for cellular PhenotypE) is an agentic framework for biological image analysis. A reasoning model drives a fixed set of imaging tools (segmentation, single-cell cropping, self-supervised representation learning, batch correction and clustering), and a verification loop decides, from label-free evidence alone, whether each tool call is good enough or has to be tuned and run again.
This organization holds the data and the run records behind the SHAPE manuscript.
| Dataset | Content | Figures |
|---|---|---|
| run-logs | Run records of every experiment shown in the main and supplemental figures: agent traces and LLM reasoning, tool calls and outputs, scores, embeddings and checkpoints, one folder per figure panel group | 1-4, S1-S4 |
| iPSC-CM-cTnT_staining | Fixed iPSC-derived cardiomyocytes imaged in brightfield, DAPI and cardiac troponin T (cTnT) | 2D-2H |
| iPSC-CM-cTnT_flowcytometry | Live iPSC-derived cardiomyocytes imaged in brightfield, with the cTnT-positive percentage of each well measured by flow cytometry | 2I-2K |
| CO_timecourse | Cardiac organoid Control vs FC time course, phase contrast and GFP at 8 time points | 3E |
| CO_drugscreen | Cardiac organoid 96-well candidate-compound screen, phase contrast and GFP at 6 time points | 3F |
Each dataset card states what the files are, where they came from and what was left out.