Predicting metabolic pathways of plant enzymes without using sequence similarity: Models from machine learning

Area of Science:

• Plant biochemistry
• Bioinformatics
• Computational biology

Background:

• Current bioinformatics tools primarily focus on enzyme function and pathway assignment using sequence similarity.
• These methods are limited for novel or uncharacterized enzymes, leading to incomplete or inaccurate metabolic pathway information.
• Enzyme annotation is crucial for understanding plant metabolism and for metabolic engineering applications.

Purpose of the Study:

• To develop a machine learning-based tool, mApLe (metabolic pathway predictor of plant enzymes), for predicting metabolic pathways of plant enzymes.
• To overcome the limitations of sequence similarity-based annotation methods.
• To enhance the accuracy and completeness of plant genomic annotations.

Main Methods:

• mApLe utilizes molecular descriptors derived from enzyme sequences for prediction.
• The tool employs machine learning models trained on these descriptors.
• It predicts metabolic pathways directly, bypassing the need for sequence similarity to reference enzymes.

Main Results:

• mApLe accurately predicts metabolic pathways for a diverse range of plant enzymes.
• The tool is effective even for enzymes lacking homologs or with incomplete Enzyme Commission (EC) numbers.
• Predictions are based on intrinsic molecular properties, not sequence homology.

Conclusions:

• mApLe offers a novel approach to enzyme annotation by focusing on metabolic pathway prediction.
• This tool can significantly improve the quality of plant genomic annotations.
• mApLe aids in identifying candidate genes for metabolic engineering research and is available online with a local GUI.