Bioinformatics Illustrations Decoded by ChatGPT: The Good, The Bad, and The Ugly

Area of Science:

• Bioinformatics
• Computational Biology
• Artificial Intelligence in Life Sciences

Background:

• Large language models (LLMs) are increasingly used for data analysis.
• LLM-based chatbots, like ChatGPT, now accept image inputs, opening new avenues for scientific interpretation.
• Evaluating LLMs for specialized scientific domains like bioinformatics is crucial.

Approach:

• Assessed ChatGPT's ability to interpret bioinformatics illustrations from cancer research.
• Evaluated performance on tasks including sequencing data analysis, drug repositioning, and tumor evolution.
• Tested the chatbot's capacity to explain plot types, apply biological knowledge, and draft figure legends.

Key Points:

• ChatGPT proficiently explains various bioinformatics plot types and integrates biological knowledge for richer interpretations.
• The model demonstrated limitations in accurately interpreting quantitative aspects of visual data.
• ChatGPT can generate figure legends and summarize findings, but requires careful verification.

Conclusions:

• LLM-based chatbots offer promising support for bioinformatics data visualization interpretation.
• Accuracy in quantitative analysis and the need for expert human oversight remain critical considerations.
• Further development is needed to enhance LLMs' reliability in complex bioinformatics figure analysis.