Predicting the pathway involvement of metabolites annotated in the MetaCyc knowledgebase.

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Metabolite associations with biochemical pathways are crucial for interpreting molecular data.
• Pathway annotations are often sparse, limiting data utility.
• Previous machine learning models predicted pathway associations using the Kyoto Encyclopedia of Genes and Genomes (KEGG).

Purpose of the Study:

• To evaluate the performance of machine learning models for predicting metabolite pathway associations using the MetaCyc database.
• To compare MetaCyc-based predictions with those from KEGG.
• To assess the utility of MetaCyc for pathway-level interpretation.

Main Methods:

• Trained and evaluated multilayer perceptron models.
• Utilized compound entries and pathway annotations from the MetaCyc database.
• Compared performance metrics (Matthews correlation coefficient) with models trained on KEGG data.

Main Results:

• Models trained on MetaCyc achieved a mean MCC of 0.845 (±0.0101).
• Models trained on KEGG achieved a mean MCC of 0.847 (±0.0098).
• MetaCyc showed a >5.6% improvement in metabolic pathway prediction performance compared to KEGG's metabolic-only pathways.

Conclusions:

• Machine learning models demonstrate comparable performance for pathway prediction using MetaCyc and KEGG.
• MetaCyc offers a more comprehensive set of pathway definitions for prediction.
• The findings support the effective use of MetaCyc for metabolite pathway association prediction.