Prompt-based bioinformatic pipeline generation for a multi-step metaviral workflow.

Area of Science:

• Bioinformatics
• Computational Biology
• Artificial Intelligence

Background:

• The increasing complexity of bioinformatics tools and multi-step analytical procedures challenges the creation of effective computational pipelines.
• Researchers, especially those with limited programming expertise, face significant hurdles in developing and maintaining these workflows.

Purpose of the Study:

• To investigate the potential of large language models (LLMs) in generating end-to-end bioinformatics pipelines.
• To evaluate the effectiveness of various LLMs in creating automated analytical workflows using a multi-step metaviral workflow as a case study.

Main Methods:

• Testing multiple large language models, including OpenAI's ChatGPT series, Anthropic's Claude series, Google Gemini, Meta Llama, and DeepSeek.
• Utilizing carefully crafted prompts and incorporating official documentation to guide LLM performance.
• Assessing pipeline generation success rates and the models' ability to handle tool substitutions.

Main Results:

• ChatGPT-4, ChatGPT-5, Claude 4.5, and Gemini 2.5 demonstrated statistically significant superior performance in generating complete bioinformatics pipelines.
• These leading LLMs effectively managed tool substitutions and benefited from prompt engineering and documentation integration.
• All tested LLMs showed potential for both initial pipeline generation and subsequent updates.

Conclusions:

• LLMs offer a promising solution for automating the construction of bioinformatics pipelines, democratizing complex analyses.
• Prompt engineering and leveraging tool documentation are key strategies for maximizing LLM effectiveness in bioinformatics pipeline generation.
• The study provides a foundation for using AI to streamline bioinformatics workflow development and maintenance.