Visualization based on the Enzyme Commission nomenclature.

Imran Shah, Lawrence HunterPubMed 1998

TL;DR

This paper introduces an interactive visualization tool that maps the Enzyme Commission (EC) nomenclature for enzyme function onto a navigable graph. This allows researchers to explore quantitative data, such as the predictability of enzyme function from protein sequence, across the entire enzyme hierarchy, facilitating hypothesis generation about prediction failures.

What Problem Does This Paper Address?

Understanding enzyme function is fundamental to biology. The Enzyme Commission (EC) nomenclature provides a hierarchical classification of enzyme reactions, but exploring the vast amount of data associated with these functions, especially quantitative metrics like prediction accuracy, can be challenging. Specifically, while sequence similarity is a common method for predicting enzyme function, its reliability varies significantly across different enzyme classes. Researchers need a way to visualize and analyze this variability to understand why certain enzymes are harder to predict and to identify patterns that might explain these difficulties.

Methods

The core of the system is an interactive graph representation of the EC nomenclature. This graph uses nodes to represent EC classes and links to show hierarchical relationships. Each node can be associated with various data attributes (stored or computed), which can then be visualized using different methods like color, size, chart types, and text. The tool allows users to interactively explore the graph, focusing on specific sub-trees or magnifying regions of interest.

A key application demonstrated is the visualization of quantitative data related to the predictability of enzyme function from protein sequence. This involves measuring the performance of sequence comparison tools (FASTA and BLAST) for predicting EC classes using the ROC statistic. The system was extended to include views for visualizing sequence alignments and interactive ROC curves.

Datasets

The primary dataset used is the EC database, specifically the ENZYME database release 21, which contains information on 1327 specific reaction types and their hierarchical classification. Protein data, including ID, name, description, and sequence, is parsed from Swiss-Prot records.

Key Insights and Ideas

  • Hierarchical Visualization for Complex Data: The EC nomenclature's inherent hierarchy is leveraged to create an intuitive visualization of complex functional relationships and associated data.
  • Bridging Sequence and Function: The tool provides a crucial link between protein sequence information and enzyme function classification, enabling the exploration of prediction reliability.
  • Identifying Causes of Prediction Failure: By visualizing performance metrics across EC classes, researchers can quickly identify enzymes that are difficult to predict from sequence and begin to hypothesize reasons, such as the presence of evolutionarily distinct subgroups or short, uninformative sequence fragments.
  • Interactive Exploration is Key: The interactive nature of the tool allows for dynamic exploration and hypothesis generation, moving beyond static data tables.

Key Results

The paper demonstrates the utility of the visualization tool by applying it to analyze the predictability of enzyme function from sequence. They measured the performance of FASTA and BLAST using the ROC statistic for 1327 EC classes. The tool enabled them to rapidly scan hundreds of functional classes and correlate sequence predictability with other attributes. For example, they identified EC 4.2.1.1 (carbonate dehydratase) as having a performance score of 0.48, with ROC analysis revealing distinct evolutionary subgroups that likely contribute to prediction difficulties. Similarly, EC 4.2.1.9 (dihydroxy-acid dehydratase) showed a performance of 0.67, with prediction issues attributed to short sequence fragments.

Limitations and Future Work

The authors acknowledge that the visualization tool is primarily designed for exploring quantitative information about EC classes. While they extended it to include sequence alignment and ROC curve visualizations, further extensions are envisioned. Future work could involve integrating this tool with metabolic pathway prediction systems to visualize known and predicted metabolic pathways, or to assist in genome sequence annotation by interactively assigning function to genes based on predicted enzyme activity. The object-oriented design of the visualization tool is highlighted as facilitating these extensions.

Summary updated

Abstract

We developed a tool for visualizing data related to protein function, based on the nomenclature defined by the International Enzyme Commission. The method represents the 1327 specific reaction types and three additional levels of abstract classifications in the nomenclature as an interactive graph. Each node and link in the graph can have associated with it stored or computed data values of various types, each of which can be assigned to any of a set of visualization methods, including color, size, various chart types and text. Visualizations can then be interactively created for single nodes, groups of nodes or the entire graph. This visualization tool is particularly useful for exploring the distribution of quantitative attributes across protein function classes. As a test of this tool, we developed a visualization of data measuring the effectiveness of sequence similarity for prediction of EC class. Using this tool, it was possible to rapidly scan hundreds of functional classes, and to correlate predictability from sequence with other attributes, facilitating the generation of hypotheses about the causes of predictive failure.
BibTeX
@inproceedings{shah1998visualization,
  author = {Imran Shah and Lawrence Hunter},
  title = {{Visualization based on the Enzyme Commission nomenclature.}},
  booktitle = {PubMed},
  year = {1998},
}
Figure