2.1.0

Code and data behind the preprint "Morpheus.AI: agents autonomously generate, execute and refine virtual tissue models" (Chandratre et al., 2026).

New in 2.1.0:

• figures/: one folder per figure and table of the preprint (Fig. 1, Fig. 2, Supplementary Figs. 1–15, Supplementary Table 1, Supplementary Note 1), each with a README, a build.py and the data and code it needs. The data are the files of the shown benchmark runs that the figures read, the per-run scores, times and costs of both final benchmarks, and the manual scores used to validate the scorer.
• compare_benchmarks.py --runs-csv figures/Fig2/data/runs_by_run.csv recomputes every statistic in the text.
• benchmark_papers/: the eleven benchmark papers published under CC BY and two bioRxiv preprints are included unchanged under their own licences (see NOTICE); the README explains how to obtain the other four.
• Preprint title, Zenodo DOI, mapping of the tool modes to setups 1–5, and corrected documentation.

The subject-facing files are unchanged from 2.0.0 (benchmark version 3). The complete benchmark run outputs are not included; they are available on request.

2.0.0

First public release of the code behind the preprint "LLM-based agents can autonomously generate, execute, and refine in silico models" (Chandratre et al., 2026).

The release contains the benchmark runner, the Morpheus skill and MCP server, the biological scoring criteria with the automated scorer (latest versions), and the analysis and figure scripts. The subject-facing files are those of benchmark version 3, which produced the preprint's main results. The README section "Reproducing the preprint" lists the software versions, the run commands and a demo that regenerates the comparison statistics from the deposited runs in about 10 s.

The paper PDFs are not included; benchmark_papers/papers.csv gives their DOIs. The benchmark runs and manual scores are archived in a separate Zenodo dataset.