Benchmark Dataset for Training Machine Learning Models to Predict the Pathway Involvement of Metabolites

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Metabolic pathways are crucial for understanding biochemical reactions, but many identified metabolites lack defined pathway involvement in public datasets.
• Existing machine learning models for metabolite pathway prediction rely on outdated and flawed datasets, hindering accurate metabolic interpretation.

Purpose of the Study:

• To develop a robust, reproducible benchmark dataset for metabolite pathway prediction using data from the Kyoto Encyclopedia of Genes and Genomes (KEGG).
• To evaluate and compare the performance of machine learning models, including Random Forest, XGBoost, and multilayer perceptron with autoencoder, for predicting metabolite pathway involvement.

Main Methods:

• Generated a new benchmark dataset from KEGG adhering to computational reproducibility standards, including updateable source code.
• Employed an atom coloring methodology and trained machine learning models (Random Forest, XGBoost, MLP with autoencoder) on the new dataset.
• Evaluated model performance using 1000 unique cross-validation folds.

Main Results:

• XGBoost binary classification models demonstrated superior performance for predicting involvement in 11 KEGG-defined pathway categories.
• Achieved a best overall weighted average F1 score of 0.8180 and a Matthews correlation coefficient of 0.7933.
• The new benchmark dataset and methodology provide a reliable resource for metabolite pathway prediction.

Conclusions:

• The developed benchmark dataset and XGBoost models significantly advance the accuracy of predicting metabolite pathway involvement.
• This work addresses limitations of previous studies by providing a reproducible and up-to-date resource for metabolic pathway analysis.
• Improved prediction accuracy facilitates deeper understanding of metabolic networks and biochemical processes.