Signature-to-Mechanisms (S2M): Benchmark Tasks for AI Mechanistic Reasoning in Biology

A collection of standardized tasks for assessing mechanistic reasoning in AI agents, created by vida-nyu. The dataset provides experimental context, molecular signatures, and prompts to test an agent's ability to reconstruct explanations from peer-reviewed biological studies. It was last updated on February 27, 2026.

Use Cases

• Benchmarking AI models' ability to infer biological mechanisms based on provided molecular signatures.
• Evaluating the quality of mechanistic explanations generated by AI agents against peer-reviewed studies.
• Training AI systems on structured tasks that formalize a core challenge in computational biology.

Strengths

• Tasks are standardized and designed specifically for evaluating mechanistic reasoning.
• Each task includes necessary elements for evaluation: experimental context, molecular signatures, and task prompts.
• Dataset is based on explanations reported in peer-reviewed biological studies.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

vida-nyu via Hugging Face

Collection Method

Formalized from peer-reviewed biological studies, but specific gathering method is not detailed.

Freshness

Last updated 2026-02-27 20:37:16; freshness should be verified.