FuncZymeDB

FuncZymeDB mines enzyme-substrate interactions from published manuscripts using an LLM-assisted workflow to improve biocuration and prediction of enzyme functions, with particular focus on plant enzymes.

Key Features:

• Large-scale literature screening: FuncFetch integrates NCBI E-Utilities, OpenAI's GPT-4, and Zotero to screen thousands of manuscripts and extract enzyme activities.
• High-precision extraction: Extensive validation revealed high precision and recall of GPT-4 in determining whether a paper's abstract indicates the presence of a characterized enzyme activity.
• Structured data extraction: Provided the manuscript, FuncFetch extracts species information, enzyme names, sequence identifiers, substrates, and products.
• Quality analysis: Extracted data are subjected to extensive quality analyses, with identified extraction errors including incorrect associations, nontarget enzymes, and hallucinations.
• Family-specific deployment: The workflow was deployed on nine large plant enzyme families, screening 26,543 papers and retrieving 32,605 entries from 5,459 selected papers.
• Comprehensive functional fingerprinting: BAHD acyltransferase activities were verified, revealing that approximately 70% of experimentally characterized enzymes are uncurated in the public domain.

Scientific Applications:

• Enzyme biocuration: FuncFetch addresses the bottleneck between published, experimentally characterized protein activities and activities deposited in databases by speeding up text-mining of protein activities.
• Substrate extraction validation: Comparison against a manually curated dataset of BAHD acyltransferase activities demonstrated a precision/recall of 0.86/0.64 in extracting substrates.
• Plant enzyme family analysis: Deployment on nine large plant enzyme families enables comprehensive functional characterization of previously uncurated enzyme activities.
• Function prediction groundwork: The workflow lays the groundwork for predicting the functions of uncharacterized enzymes by establishing comprehensive functional fingerprints.

Methodology:

FuncFetch integrates NCBI E-Utilities, OpenAI's GPT-4, and Zotero to screen manuscripts and extract enzyme activities. GPT-4 determines whether a paper's abstract indicates characterized enzyme activity, then extracts species information, enzyme names, sequence identifiers, substrates, and products from selected manuscripts, with extracted data subjected to extensive quality analyses.