Metabolite and reaction inference based on enzyme specificities.

Area of Science:

• Biochemistry
• Systems Biology
• Bioinformatics

Background:

• Enzymes often exhibit promiscuity, catalyzing reactions beyond their annotated functions, information typically found in specialized databases like BRENDA.
• Predicting enzyme aspecificity is crucial for expanding knowledge of metabolic pathways and identifying unknown biochemical transformations.

Purpose of the Study:

• To develop a computational model for predicting enzyme aspecificity.
• To enable the prediction of whether a given compound can be transformed by a specific enzyme.

Main Methods:

• Developed the metabolite and reaction inference system based on enzyme specificities (MaRIboES).
• Utilized structural and stereochemistry similarity measures, along with molecular fingerprints, to generalize enzymatic reactions.
• Employed data from the BRENDA enzyme activity database for model training and validation.

Main Results:

• Achieved 80% accuracy in predicting known enzymatic reactions using leave-one-out cross-validation.
• Applied MaRIboES to yeast glycolytic and pentose phosphate pathways, predicting numerous known and novel reactions.
• Identified potential metabolic bypasses and cross-links, including the formation of novel compounds.

Conclusions:

• The MaRIboES system effectively models enzyme aspecificity, expanding the understanding of enzymatic capabilities.
• This predictive capability has significant applications in completing metabolic networks, metabolic engineering, and mass spectrometry data analysis.
• The freely available Matlab and C++ code facilitates further research and application in systems biology.