Data science through natural language with ChatGPT's Code Interpreter.

English summary

This article describes the potential of ChatGPT’s Code Interpreter, which can write and execute code, to support biomedical researchers through natural-language interaction. Conversational workflows can cover data loading and exploration, model development and comparison, permutation feature importance, partial dependence plots, and additional analyses and recommendations, potentially allowing researchers to focus more on higher-level aspects of their work.

Key points

• Large language models (LLMs) are presented as tools for biomedical research because of their ability to understand and generate human-like text.
• Code Interpreter links natural-language interaction with code writing and execution, helping streamline data-analysis workflows.
• Using materials from a previously published tutorial, users can conduct several types of analysis through conversation with the chatbot, including data exploration, model development and comparison, permutation feature importance, and partial dependence analysis.

Audience

Biomedical researchers, data scientists, medical and pharmaceutical researchers, and educators involved in data-science training.

Scope and limitations

The supplied text provides a general discussion of the potential uses and concerns of LLMs and ChatGPT’s Code Interpreter; it does not report a specific dataset, model-performance metrics, or quantitative comparisons. Therefore, accuracy, reproducibility, and generalizability of actual analyses cannot be assessed from this text alone. The limitations and concerns identified in the source include the need for critical thinking, privacy, security, equitable access, and appropriate human supervision and responsibility.

Research context

The topic has a thematic connection to Sangzin Ahn’s portfolio through natural-language analytical tools, research data science, and the use of artificial intelligence in biomedical research.