Benchmarking Large Language Models for Transcriptomic Data Analysis and Visualization: Multi-Model Workflow Evaluation

Prompt specificity improves code structure and methodological correctness in LLM- and agentic-coding workflows for transcriptomic COVID-19 response analysis, but quantitative performance varies with preprocessing and execution environment.

• Why it matters: Automated analysis and visualization of transcriptomic data needs reliable, reproducible workflow components, yet current LLM capabilities for scientific targets remain uncertain.
• What they did: ChatGPT 5.5, Claude Sonnet 4.6, Gemini 3.1, and Llama4-Scout were benchmarked on scientific prompts covering dimensionality reduction, clustering, outlier detection, reproducibility, and code generation, with both quantitative metrics and qualitative code checks.
• The result: Quantitative results depend heavily on preprocessing choices and execution environment, and alignment with label-based clustering targets was inconsistent, indicating promise alongside limitations for transcriptomic workflow automation.