SCIGYM
An agentic systems-biology suite in which language models iteratively perturb simulated SBML systems, analyze time-series observations in Python, and reconstruct hidden biological reactions.
Scientific task · Systems modeling
Simulation-based experiment
Interact with or reason over a simulator to identify
仿真实验
Definition and search aliases
- Permanent ID
simulation-based-experiment- Aliases
- interactive simulation, system identification
- Deprecated aliases
- None
- Hierarchy
- Leaf task under Scientific workflow and systems analysis
Coverage
1 benchmark families cover this task
SCIGYM Large
The formally released SCIGYM track containing the 213 systems not included in the creator paper's model evaluation, with systems reaching up to 400 reactions.
SCIGYM Small
The formally released and creator-evaluated SCIGYM track containing biological systems with fewer than ten reactions.
Evidence-backed count claims
Each row keeps its original unit and basis. Rows with different units or overlapping mappings are never added.
| Benchmark | Mapped task | Coverage | Count | Version | Evidence |
|---|---|---|---|---|---|
| SCIGYM root: scigym | Simulation-based experiment official-taxonomy · high | explicitly-in-scope | 350 systems distinct curated BioModels systems released as SBML benchmark instances | 2025 release | scigym-evidence-release-countsscigym-evidence-taxonomy |
| SCIGYM Large root: scigym | Simulation-based experiment official-track · high | explicitly-in-scope | 213 systems unique SBML systems in the official large Parquet split, containing the remaining systems with up to 400 reactions | 2025 release | scigym-large-evidence-count |
| SCIGYM Small root: scigym | Simulation-based experiment official-track · high | explicitly-in-scope | 137 systems unique SBML systems with fewer than 10 reactions in the official small Parquet split | 2025 release | scigym-small-evidence-count |
Official evaluations connected to these benchmarks
Runs are included only for benchmark records mapped here (and formal child tracks when a mapped suite is the root). A task mapping does not imply that every run isolates this task.
| Work | Provider / class | Related runs |
|---|---|---|
| Measuring Scientific Capabilities of Language Models with a Systems Biology Dry Lab | University of Toronto, SickKids, Axiom, Mila, Vector Institute benchmark_creator | scigym-small-creator-paperscigym-small-zero-shot |