Multimodal knowledge expansion widget powered by plant protein phosphorylation database and ChatGPT

Area of Science:

• Bioinformatics
• Plant Biology
• Artificial Intelligence

Background:

• Biological databases have rigid structures limiting user exploration.
• Plant phosphorylation data is vast but underrepresented in curated databases.
• Integrating large language models (LLMs) into bioinformatics tools is challenging.

Purpose of the Study:

• To develop a multimodal question-answering widget integrating ChatGPT-4o with the Plant Protein Phosphorylation Database (P3DB).
• To enable natural language querying and visual data extraction from scientific figures for plant phosphorylation research.
• To enhance user accessibility and extend the utility of biological knowledgebases.

Main Methods:

• Developed a multimodal question-answering widget connecting ChatGPT-4o with P3DB.
• Implemented natural language query support and dynamic prompt formulation.
• Explored prompt strategies (one-step, two-step, few-shot, image-cropping) for optimal accuracy.

Main Results:

• The ChatGPT-P3DB integration supports exploration of phosphorylation events, kinase-substrate relationships, and protein-protein interactions.
• The widget successfully extracts regulatory pathways and phosphorylation markers from scientific figures using image interpretation.
• Tested multimodal LLMs (ChatGPT-5, Gemini 1.5) showing comparable capabilities.

Conclusions:

• The ChatGPT-P3DB integration enhances user accessibility to plant phosphorylation data.
• The system enables visual data extraction from complex figures, extending knowledgebase utility.
• The developed framework offers a flexible and adaptive approach for biological knowledge discovery.