AI Virtual Cell Model Predictions and Evaluation Metrics for Scientific Discovery

Model predictions and evaluation metrics supporting the paper "Are Current AI Virtual Cell Models Useful for Scientific Discovery?". The 10.0 GB dataset includes predicted gene-level perturbation effects and reasoning traces from large language models. Author Michael Bereket released the data on figshare in May 2026 to enable reproduction of the study's figures.

Use Cases

• Reproducing evaluation figures from the associated paper based on the provided metrics.
• Benchmarking AI virtual cell models against the predicted perturbation effects.
• Analyzing the reasoning traces of large language models used in the study.
• Investigating gene-level and gene-set-level effects of perturbations for biological discovery.

Strengths

• Dataset size is 10.0 GB, indicating substantial analysis data.
• Includes multiple data components: predicted effects, evaluation metrics, and LLM reasoning traces.
• Explicitly linked to a peer-reviewed preprint on bioRxiv and associated GitHub code for reproducibility.
• Released under the permissive CC-BY-4.0 license.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect methodological or source bias inherent to the specific models and analyses in the original study.

Provenance

Source

figshare, associated with the paper "Are Current AI Virtual Cell Models Useful for Scientific Discovery?"

Collection Method

Likely generated from AI model predictions and subsequent evaluation as part of the described research.

Freshness

Last updated 2026-05-01 21:08:13